Microsoft-Decision-1: The System One Engine for Probabilistic JSON States

Introduction

Architecting highly resilient computing platforms inherently demands specialized components capable of processing enormous queues of transactional decisions in a flash with impeccable precision.. As long as the assessment tool perfectly fits the big enterprise environment, collaborating tightly with analysis and content creation modules, then the whole system works like a charm without any technical snags. Naturally, this approach helps to considerably increase the effectiveness of self-guided algorithms and intelligent operation assistants, so that complicated processes are running with continuous speed without any delays.

Going through the most recent cloud environment launches will provide plenty of reasons to explain the popularity of the Decision-Microsoft-1 tool among system engineers. The module ensures an indispensable insurance mechanism for sophisticated enterprise logic by transforming unclear information into definite triggers for actions while strictly taking care about the compatibility of those processes with the limits set by the company policies. Comprehending the importance of such reliable routing mechanism is a first step towards understanding its inner mechanisms, wide-range applicability in the large corporate settings, and certain precautions connected with its implementation. Analyzing comparative advantages, which the framework offers, as well as existing channels of its business implementation, enables technology leaders with the exact blueprint needed to architect next-generation solutions.

What is Microsoft-Decision-1?

Microsoft-Decision-1 is a custom trained decision scoring model designed for scoring unstructured text or JSON inputs and giving calibrated probabilities over fixed options without generating natural language text. Acting naturally as a System One layer of intelligent applications, it takes bounded situations and gives typed numeric decisions, thus decreasing the computation costs compared to generic conversational models.

Key Features of Microsoft-Decision-1

• 32K Token Context Window for Single-Pass Scoring: Handles gargantuan input states up to 32,768 tokens in a single pass. It structurally outperforms existing decision architectures that usually limit the number of options to 255 in a single pass and require complex two-pass routing processes.
• Absolute Perturbation Robustness & Zero Decision Flips: Demonstrated robustness in rigorous testing using eight different perturbation factors (including formatting perturbation and key renames), changing its decisions in only 1.3% of noise-injected cases. Importantly, it demonstrated a perfect 0% decision flips rate in cases where options were paraphrased, reversed, and reordered.
• Structured Decision Primitives with Native Abstention: Includes natively three different decision primitives: noul (boolean values), choice (multi-label classification), and score (ordinal scoring rubrics). Unlike traditional binary classifiers, it provides an explicit option for abstention (e.g., “can’t tell”) to completely eliminate any forced false classification in case of insufficient evidence.
• Data Zone Sovereignty (DataZoneStandard): Offered only within the geographic limits set (US Datazone & EU Datazone) to ensure that all inference computations conform to compliance requirements in the region.
• Enterprise Security & Governance Integration: Seamlessly integrates with the organization’s framework through Microsoft Entra ID (OAuth2 Bearer tokens), making use of the granular Azure RBAC architecture along with single Azure billing subscription.

Use Cases of Microsoft-Decision-1

• High-Speed Player Feedback Sentiment Categorization: Provides the ability to classify enormous numbers of player feedbacks and reviews from different sources such as Steam, X (formally called Twitter) and surveys into precisely structured categories. As it bypasses the significant computational cost associated with generative models, studios will be able to process world-wide player feedbacks instantly which will increase the speed of releasing game patches and improve experiences for the players.
• Science Adaptive Replan Loops (Microsoft Discovery): It works as an ultra-consistent rubric grader of long and iterative loops of scientific experiments. In these kinds of environment, automated agents perform tasks on analyzing previous experimental data, refining their hypothesis, and doing further experiments. As it provides a high-speed grading system without delay of old systems, it is helpful in reducing R&D discovery timelines and physical experiment wastes in laboratories.
• Copilot Response Quality Control Real-Time: It works as an ongoing architectural safety check by judging secondary agentic responses of the system to user chats according to quality and groundedness rubrics before sending back to the end-user. It helps in verifying responses in real time and maintaining very rigorous user experiences without making any UI delays for the enterprise applications.
• On-Call Engineering Incident Knowledge Retrieval: Helps in providing a very crucial filter layer during real-time system outage by quickly searching, filtering, and prioritizing the operational knowledge that exists across various disjointed engineering systems like tickets, chat channels, live call centers, etc. The quick filtering of such context helps in reducing the mean time to resolution for the problem thus ensuring system availability.

How Does Microsoft-Decision-1 Work?

Microsoft-Decision-1 deviates from traditional text-generation methods and concentrates solely on the probabilistic JSON states returned. At first, this project was designed by training the comprehensive full-base model after post-training on Alibaba's openly released weights of Qwen3.5-9B. Instead of using a frozen backbone and doing partial optimization, this framework chose to go for full-parameter optimization. In contrast to other contrastive architectures such as CLM-8B, which trains only a 20 million parameters MLP projection head using a frozen LLM encoder based on InfoNCE loss, this special model is fine-tuned for the entire neural weights structure. The process is augmented with safety fine-tuning that enforces refusal boundaries in a total of 5,250 requests through 11 safety benchmarks.

source - https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/introducing-microsoft-decision-1-in-microsoft-foundry-for-decision-and-classific/4562742

In terms of operation, the engine communicates through an API channel tailored specifically to receive structured state inputs, as well as typed question primitives. Moreover, its design is such that its roadmap is independent of any particular architecture. Although its current release depends on the Qwen3.5-9B foundation model, it does not alter its deployment contract and pipelines at all. In fact, as the system gets upgraded to more powerful Microsoft AI (MAI) or OpenAI foundation models, there is no need for making any changes to its integration code.

Performance Evaluation with Other Models

In terms of quantitative analysis, performance evaluation of specialized classification engines is done by separating pure logic from prose generation. Testing was done using the extensive 36-Benchmark Evaluation Suite as shown in table below, which consists of 147,137 blind questions in which the engine obtained the highest overall ranking. It outperformed other decision and generation models such as Jev 1.13.0, Quyet-1.0-Large, Surogate Rune 26B-A4B, deck-31B, and H2O-Lightning-4B. The significance of this feat ensures that the engine really learns the decision-making logic on unseen data and does not merely memorize the texts. In line with this, the JevBench Speed Test separated pure inference speed. Here, it worked much faster than the second-placed H2O-Lightning-4B v1.1, and even more so compared to GPT-6 Sol, showing that efficient routing mechanisms can significantly lower the architecture's latency.

source - https://commandline.microsoft.com/microsoft-decision-1-model-foundry/

In terms of infrastructure-level and real-world metrics, the Foundry Latency Benchmark demonstrates an operational profile designed specifically for high availability systems. In this case, the engine showed strong competition against and beat legacy reasoning APIs thanks to highly constrained median and tail-end latency thresholds, making it possible for developers to fit several classification gates in one user request. Finally, the Internal Production Benchmark for Data Labeling compared the performance of the system to that of massive organizational datasets. It proved itself equal to qualitative results of legacy titans such as GPT-6 Sol, but at lower prices and much faster. Thus, it was shown conclusively that using a massive reasoning engine for basic classification purposes is inefficient.

How to Access and Use Microsoft-Decision-1?

Technical users, such as engineers and infrastructure operators, can instantiate Microsoft-Decision-1 through the Microsoft Foundry Model Catalog or use the remote endpoints of Microsoft-Decision-1 through the OpenRouter platform. In spite of being trained on an open weight model, Microsoft-Decision-1 deployment is entirely commercial cloud-based and not local weights deployment but only networked model calls are possible. Its operation implies having an active Azure subscription and valid payment means, having access to Microsoft Foundry projects, and having a model management role, i.e., the Cognitive Services Contributor RBAC role in Azure.

Limitations 

Notwithstanding its exceptional accuracy, the model comes with rigorous limitations of operation: it is absolutely forbidden to deploy it as the single automated decision-maker in high-stake fields where there are serious consequences for people – such as hiring, lending, housing, health care, insurance or any other legal right where obligatory human review is required. Other limitations may include the risks associated with sensitivity to specific question formulation or option ordering, the possibility of training data bias reproduction and the risk of false positive/negative content filtering results.

Future Plans 

In order to resolve the issues outlined above in future instances, the development roadmap includes the rebase onto proprietary MAI and OpenAI platforms, taking into account continuous feedback from the datasets for the improvement of probability calibration, reduction of adversarial blind spots and increase of cost effectiveness.

Conclusion

While the demand for more efficient and consistent computational performance becomes stronger, the treatment of every automated interaction as a generative prompt has become an extremely dangerous anti-pattern from the architectural perspective. Microsoft-Decision-1 demonstrates an important step forward in the field of deployment strategy, being the final foundation 'System One' layer for intelligent systems. For all those who manage complex systems, the key value of the solution is obvious. 

Sources
https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/introducing-microsoft-decision-1-in-microsoft-foundry-for-decision-and-classific/4562742
https://commandline.microsoft.com/microsoft-decision-1-model-foundry/
https://ai.azure.com/catalog/models/Microsoft-Decision-1
https://learn.microsoft.com/en-us/azure/foundry/foundry-models/how-to/use-foundry-models-microsoft-decision

Disclaimer - This article is intended purely for informational purposes. It is not sponsored or endorsed by any company or organization, nor does it serve as an advertisement or promotion for any product or service. All information presented is based on publicly available resources and is subject to change. Readers are encouraged to conduct their own research and due diligence.

How Mistral Large 4 OpenWeight MoE AI Tops GPT-6 Astra Cyber

Introduction

A technological workload mandates a multimodal architecture that goes beyond surface-level perception in order to perform related tasks within complex parsing of the visual layout structure, program development through multiple levels, formal numerical computation, and security auditing procedures. The necessity for the connection of all of these features into one single architecture lies in the fact that, typically, engineering or analysis challenges never consist of just perceptual processing, but of immediately following up this process by performing an execution of codes, computing of equations, or auditing security without latency.

All of the mentioned features can be achieved in Mistral Large 4 through connecting a native vision encoder with 1.6 billion parameters with a 1-trillion-parameter sparse Mixture-of-Experts architecture (with ~52 billion parameters active per token), and a context window of up to 1 million tokens. In contrast to commercially closed solutions, where security-critical information flows via external endpoints and is being subject to conservative refusal filtering that blocks any vulnerability auditing, Mistral Large 4 allows for frontier visual grounding and multi-stage symbolic reasoning.

What is Mistral Large 4?

Mistral Large 4 is an advanced multimodal model that employs the fine-grained Mixture-of-Experts (MoE) architecture. It combines deep instruction following, mathematical reasoning, visual perception, and agentic tool execution in one unified sparse weight configuration.

Key Features of Mistral Large 4

• Sovereign Cybersecurity and Censor-Free Defensive Utility: Combines superior defensive cyber skills with open-weight self-deployment, enabling cybersecurity professionals to test, reproduce, and fix real-life vulnerabilities without facing fake safety restrictions or too much over-refusal from the provider.
• Precise Visual Grounding and Technical Perception: Includes a specialized vision encoder with 1.6 Billion parameters that can be used to analyze gigapixel satellite images to assist in disaster recovery and dissect complex computer-aided designs (CAD) all the way to their mechanical parts.
• Multilingual Support: Fluently understands more than 160 native languages, including each language used in the European Union.
• Proficient in Enterprise Legal and Financial Domain: Very specialized in legal document synthesis, tracking compliance, and exploring corporate financial databases step by step.

Use Cases of Mistral Large 4

• Secure Financial Research and Trading Strategies Backtesting: Financial firms can utilize this technology and directly apply the model onto their existing infrastructure. In this way, the analysis of very large, constantly changing data sets like SEC filings and regional company statements will be conducted. As all processing occurs on-site, financial research desk can run parallel automated valuation and backtest private strategies without any need of revealing very sensitive actions by using third-party cloud services' API.
  
• Private Agent Building and Orchestration: Machine learning engineers inside the firm will be able to build custom, domain-tailored autonomous agents in-house. Since the system provides support of offline terminal and sandbox orchestration with reinforcement learning environment, companies will be able to precisely control agent alignment and training loop, ensuring that their bespoke artificial intelligence works exactly as required without any additional training setup from external resources.
  
• Software Adaptation and Private Source Code: Development teams expanding their operations into other regions are able to automate the process of translation of complex technical documents, localization of user interface and context-aware code refactoring. In this way, developers will be able to easily adapt their software for hundreds of native dialects without exposing their private intellectual property and code that remain securely locked inside the corporate firewall.
  
• Sovereign Cross-Platform Workflow Automation: Operational managers in heavily regulated industries like healthcare, government, and finance can automate workflows that easily connect across common enterprise software such as email clients, spreadsheets, and CRM systems. This results in extensive workflow automation at predictable compute costs, all while maintaining complete control over critical data and complying with regional requirements for digital sovereignty.

How Does Mistral Large 4 Work?

Mistral Large 4 uses a very large 1.05 T parameter model size based on a Mixture-of-Experts design which activates from 49 to 52 billion parameters per token. Granular routing in this model is combined with a vision encoder with 1.6 B parameters and 1-M-token context window, optimized for commercial workloads by utilizing efficient caching and serving economics. Training pipeline of the model is done using a cluster of 3,800 NVIDIA Grace Blackwell GPUs which are located in European data centers. 

Continuous reinforcement learning produces about 33B tokens each day and produces about 16B trainable completion tokens after strict filtering and loss masking. Async autoscaling fleet of actor workers produces many parallel rollouts which allows doing long trajectory optimization in different context compactions without off-policy drift. Moreover, single-turn chat, scientific problem solving, safety alignment and factuality are done in a unified reinforcement learning run together with live code execution sandboxes, web search and external APIs with a composable verification suite consisting of unit tests, reward models, static checks, LLM judges similar to those provided by Mistral Forge.

Performance Evaluation with Other Models

The Mistral Large 4 has shown exceptional superiority in benchmarks regarding cybersecurity and safety. The architecture shows top results on vulnerability reproduction and Cybench security evaluations. Superior to commercial models with restrictive refusal filters, such as Claude Opus 5.5 and GPT-6 Astra, which demonstrate sharp performance decline because of those restrictive mechanisms, the architecture shows the compatibility between unrestricted execution mode and prompt-injection resistant behavior on evaluations like Lakera B3 and KORA and earns excellent results as an open-weight system.

source - https://mistral.ai/news/mistral-large-4/

The architecture has shown great success in benchmarks related to agentic coding and terminal work flows. Specifically, the benchmarks include the DeepSWE, SWE-Atlas-QnA, and Terminal-Bench tests. In the Combined Coding Agent Index test, where it surpasses DeepSeek V4 Pro and Qwen3.8 Max architectures, the architecture shows good multi-step code repository navigating and patch generation abilities, performing exceptionally well in human-evaluated test tasks.

source - https://mistral.ai/news/mistral-large-4/

In enterprise knowledge work, legal synthesis, and multimodal perception, further assessments show better results on AutomationBench and the HarveyAI Legal Agent benchmark. The model shows the best performance in document intelligence, complex chart evidence extraction, chart navigation, and multi-step financial database querying on FinWorkBench, surpassing other models in applied mathematics, theoretical physics reasoning, and quantum simulation accuracy while keeping the highest visual grounding scores on dense perception benchmarks.

How to Access and Use Mistral Large 4?

Mistral Large 4 is now available for public preview through Mistral Studio API. Users can interact with the model in an interactive playground and programmatically utilize it through various standard endpoints including chat completions, multi-turn conversations, batch inference, and agentic use cases. In case of need for a self-hosted deployment, the official open model weights will be available by the end of the month and users will be able to run it in their private cloud or on-premises to ensure complete data sovereignty, secure security operations, and offline capability. There is also an extended version of the model which has less moderation and increased cyber utility before the public weight release for selected enterprise customers, cybersecurity experts, and state entities involved in red teaming.

Potential Future Enhancements

Going forward, the open-weight accessibility of this model opens up the potential for structural evolution separate from typical capability upgrades. In terms of cybersecurity, an avenue may present itself for implementing zero-knowledge proof generation into the process of automated patching. By doing so, the machine will be able to cryptographically validate that certain vulnerability patches have been executed on secure networks without revealing any of the details regarding their exploit. Concerning the sphere of software development, there are opportunities for the development of technical debt prediction through inference.

In the fields of advanced mathematics and multimodal document analysis, there exists the possibility of achieving a cross-dimensional reconstruction. The visual grounding features might be used to extract the topological information from 2D drawings like engineering patents and regulatory blueprints in order to recreate them in interactive 3D environments for compliance and mechanical stress testing purposes. The same goes for the mathematical execution paths – they might not only be used for solving equations but manipulating simulated physical constants during materials discovery research.

Conclusion

The Mistral Large 4 shows that absolute data sovereignty of enterprises along with frontier-level technical capability can be effectively combined. By getting rid of artificial limits to cybersecurity, multimodal perception, and long-horizon reasoning, this model provides technical leaders with the necessary autonomous capabilities to create resilient and self-hosted AI infrastructure for the next decade.

Sources:
https://mistral.ai/news/mistral-large-4/
https://docs.mistral.ai/models/mistral-large-4-0

Disclaimer - This article is intended purely for informational purposes. It is not sponsored or endorsed by any company or organization, nor does it serve as an advertisement or promotion for any product or service. All information presented is based on publicly available resources and is subject to change. Readers are encouraged to conduct their own research and due diligence.

CLM-8B: Dual-Encoder Scoring and Agentic Trajectory Verification

Introduction

Recent studies in artificial intelligence are making more use of the technique of contrastive representation learning that enables the joining of contextual situational knowledge and subsequent operational activities. The direct linking of states and actions into a continuous vector space is made possible by Contrastive Language Models. The latter provides an overall approach for the evaluation of alternatives through semantic matching of decisions in continuous vectors. This enables the operational processes to be significantly faster when running, especially while implementing difficult tasks such as programming or gaming simulation. In addition, with clean scalability of vector alignment relative to data and computing power, the increase in computing and data resources leads to higher precision in the system.

For those seeking a flexible open-weights approach to contrastive learning, CLM-8B represents the ideal baseline model for modern decision-making. This model offers practical ways of implementing the evaluation of candidates and their trajectory using contrastive projections on a well-established foundation model. Using CLM-8B helps to create a flexible architecture that does away with proprietary operational dependencies in the process of deployment.

What is CLM-8B?

CLM-8B is a joint effort between the researchers from Stanford University and NVIDIA Research. This is an 8B parameter contrastive language model that serves as an open-weights 'System One' decision engine. Whereas traditional generative language models use auto-regressive token sequence prediction, CLM-8B is capable of evaluating and ranking actions based on the cosine similarity of their 512-dimensional vector representations relative to incoming environmental state.

Key Features of CLM-8B

• Independent Action Vector Caching: Precomputes and caches embeddings of potential actions  to GPU directly, needing only one pass over newly arriving environmental states.
• Zero-Token Cost for Cached Vectors: Metering  strictly counts only encoder tokens consumed by missing vectors in cache, therefore using cached vectors of actions or states takes zero encoder tokens.
• Wire-Format TypeSafe Primitive Availability: Ensures complete wire-format availability for decision-making primitives such as Noul for probability calibrated Boolean decisions, Choice for criterion multi-option decisions, and Score for rubric scoring, thus enabling direct replay of TypeSafe/Jev requests.
• Candidate Reranking Primitive: Provides a native endpoint and API-method  for evaluating and ranking arbitrary candidate text lists in one pass.
• Robotics Extension (CoVer-VLA): Applies the contrastive state-action verification concept to the robotics setting where VLA models can validate their physical trajectory via the CoVer-VLA framework.
• Prose Plain Language State Representation: Encodes state information as prose text (key: value for dictionary and -item lists for arrays) instead of encoding states as a string of JSON which skips completely parsing and schema matching issues.

Use Cases of CLM-8B

• Cost-Free Memory Management for Repeated Environment Navigation: Autonomous systems regularly navigate consistent digital environments, admin portals, or standard operational processes. CLM-8B maintains memory management cost efficiency by storing the representations of previous decisions of a model. This removes the need to incur the recurrent token cost calculation when navigating a known UI state, while keeping a steady VRAM memory footprint that ensures stability of servers during a long-running process.
• Low-Cost Adaptation for Proprietary Enterprise Taxonomies: Enterprises often have the need for an automated decision model that is aligned with their internal processes, special product catalogs or proprietary taxonomies of operation. CLM-8B supports such low-cost adaption in enterprise specific datasets with only updates to their lightweight projection heads of 20 million parameters.
• Scalable Action Routing for Large Candidate Spaces: The digital workflows of agents can often have hundreds or thousands of candidate tools as options, such as extensive microservice collections or databases in e-commerce. CLM-8B scales well in dealing with large candidate spaces by separating state evaluation from candidate embedding. This allows enterprises to scale their capability of integrating new tools without facing latency issues or cost explosion.

How Does CLM-8B Work?

The architecture of CLM-8B utilizes the concept of a dual-encoder which is constructed based on a frozen language model backbone. Instead of training the model of eight billion parameters from scratch or doing fine-tuning of the whole set of parameters, the system utilizes a frozen Qwen3-8B base encoder which extracts the hidden states representations of the input text. Two lightweight encoders with 20 million parameters—each dedicated for either environmental states or candidate actions—are used with the frozen base model. The projection heads convert the high-dimensional output from the base encoder into normalized continuous vector embeddings. While inferring, the system computes the similarity between the state embedding and each candidate action embedding. Then the system computes the scaled relative probability distribution over these similarity scores to get the ranking of the optimal action choice without generating text token-by-token.

source - |https://contrastive-lm.notion.site

This strong verification capability is achieved through a three-step alignment training process. At the initial stage of pre-training, the projection heads are trained on about sixty million question-answer pairs in order to lay down a well-defined decision space. At the second stage of mid-training, about thirty million artificial hard negatives are added to make the system more discriminative among similar and plausible choices. At the third and final stage of post-training, the system is trained on one million complex execution paths of agents. In order not to forget earlier learned concepts, this stage includes mixing of new trajectory data with replayed earlier trajectory data.

Performance Evaluation with Other Models

When considered as a trajectory verifier for multi-step agentic coding tasks, fine-tuned CLM-8B sets new records in comparison with other decision engines. On the DeepSWE benchmark, the fine-tuned CLM-8B showed 81.6% verification accuracy over 38 held-out tasks when choosing the best-of-N solution candidates produced by frontier models such as Opus 5. At the same time, proprietary joint-evaluation architectures like TypeSafe's Jev did not work as verifiers on this benchmark as their performance was worse than the random selection Pass@1 baseline.

source - |https://contrastive-lm.notion.site/

On the Terminal-Bench 2.1 benchmark, the fine-tuned CLM-8B showed 87.6% accuracy over 30 held-out tasks when verifying multi-step shell execution traces produced by Fable 5. Importantly, CLM-8B provided these verification decisions 4.1× to 5.7× faster than joint autoregressive forward passes on NVIDIA H100 GPUs. This proves that the decoupled contrastive scoring is efficient.

For short-horizon tasks such as browser games  and game scenarios, CLM-8B is on par with Jevs in terms of decision correctness with up to 9× lower average latency. With candidate spaces growing to around 1,000 actions, CLM-8B shows a 13× speed-up compared to uncached joint model evaluation thanks to the caching action vectors capability. Task-specific fine-tuning provides state-of-the-art verifiers for long-horizon executions.

How to Access And Use CLM-8B

CLM-8B is hosted on the GitHub platform in the repository named Contrastive-LM/CLM where the model parameters can be found on Hugging Face and documentation on the Contrastive-LM Notion webpage. The model can be self-hosted locally by pairing it with a local embedding server, which involves a local web playground along with the Python SDK. The CLM projection head checkpoints along with the Qwen3-8B base encoder are released as per the open-source Apache 2.0 license and can thus be used freely for any commercial purpose.

Limitations

The CLM-8B model has a number of significant limitations. First, the architecture is encoder-locked because the twenty-million parameter projection heads used by CLM-8B rely only on last-token-pooled embeddings generated by the Qwen3-8B base model and cannot be used with other backbones. Second, the model does not generate freeform text, relying solely on scoring and ranking provided candidate options. Finally, the ability of the model to generalize in zero-shot open-domain mode is limited by the eight billion parameter count. The future work will consist of developing CLM-35B in the multimodal domain.

Conclusion

CLM-8B represents a design shift in developing agents' infrastructure, which is changing the token-by-token generative evaluation for continuous vector-space contrastive alignment of the System One. The developers of scalable agentic infrastructure can use CLM-8B as a low-overhead, weights-open architecture, allowing the decision routing to be computationally and monetarily feasible.

Sources:
|https://contrastive-lm.notion.site/
https://github.com/Contrastive-LM/CLM
https://huggingface.co/Contrastive-LM/CLM-v0.1-8B

Disclaimer - This article is intended purely for informational purposes. It is not sponsored or endorsed by any company or organization, nor does it serve as an advertisement or promotion for any product or service. All information presented is based on publicly available resources and is subject to change. Readers are encouraged to conduct their own research and due diligence.

Why MiMo-V2.6 Pro Defeats Claude Fable 5 In DeepSWE v1.1

Introduction

Open-weights artificial intelligence is now at a crossroads where getting more intelligence for money determines the feasibility of production. The development of native omnimodal architectures together with scalable reinforcement learning is a recipe for establishing a transparent environment for model improvement. Making a foundation model work as an engine of both creation and analysis makes it possible for modern systems to merge textual reasoning with spatial, musical, visual, and code execution. What is essential about such models is that using groupwise agentic grading gives a way to get sophisticated reward signals, which help models to tune their complicated tool-using strategies without getting stuck in binary reward schemes. Alongside with using special optimization algorithms to guarantee training stability, open-weights models deliver frontier execution for much lower costs than those of proprietary APIs.

MiMo-V2.6 is one of the milestones in this direction – it occupies the top of world leaderboards among open-source projects and also has affordable API prices. This technology presents a single unified baseline that combines post-training reinforcement learning, open environment verifiers, and native multimodal perception.

What is MiMo-V2.6?

MiMo-V2.6 refers to an omni-native sparse mixture-of-experts (MoE) foundation model family that is purposefully designed to advance the frontier of intelligence based on inference and training costs. The model can process text, high-res images, videos, and raw audio inside a single context window of 1M tokens. MiMo-V2.6 uses reinforcement learning with large compute on multiple domains in agent-based environments to perform end-to-end tasks.

Model Variants

• MiMo-V2.6-Pro / Pro-RL: The premier MoE model with 1.02T parameter count in total with 42B parameters per token with routing to 384 experts (8 experts per token). Structured with 70 layers (one dense layer initially and 69 MoE blocks) and hidden dimension of 6,144, it has been designed for deep reasoning and long-horizon software engineering. It can be downloaded in safetensors form from a 524GB file after spending $2.62M on post-training RL compute.
• MiMo-V2.6-Flash / Flash-RL: The efficiency-balanced MoE variant featuring 309B/310B total parameters with 15 Billion active parameters per token across 256 routed experts (8 active per token). Built with 48 Transformer layers (1 dense + 47 MoE blocks) and a hidden dimension of 4,096, it delivers near-flagship agentic performance at a significantly reduced compute footprint ($0.85M RL post-training cost) and is available as a 159GB safetensors file.
• MiMo-V2.6-Distill-Qwen-9B: A compact 9B parameter dense image-text-to-text model fine-tuned from Qwen3.5-9B via Supervised Fine-Tuning (SFT) on 77.4 Billion synthetic tokens (27.2B loss-bearing tokens) generated directly by MiMo-V2.6. Balanced across coding , general agents, visual coding, and cybersecurity, it brings high-efficiency agentic capabilities to resource-constrained edge deployments.

Use Cases of MiMo-V2.6

• Making Animated 3D Scenes & Working with Robots: It can convert any type of written text, photo, or video into a 3D object, something functional in Blender. This is achieved by simplifying complex modeling and programming processes associated with robotics and gives one a real-time feedback on things happening to robots in 3D.
• Music Composition, UI Design & Multimedia Editing: It performs music composition and produces MIDI files people can work with on DAWs, produces UI design screens via Figma, and voice recordings. All of this make the creative work easier and gives independent creators a chance to compete with major companies by taking control of their audio-visual content.
• Materials development and theoretical proofs validation: Reads complex patents and literature and configures its computer programs in order to search applications of materials in the green industry while acting with active agents and generating codes for formal proofs of mathematical theorems—this enables minimizing costs of laboratory testing and patent-approving processes for scientific researchers and obtaining a pulpit for formal proofs generator.
• High Throughput Production Hosting Using Block Diffusion Speculation: Makes use of the block diffusion speculative decode along with the optimization of the inference engine in order to considerably increase the speed of the production of the output without compromising on accuracy.
• Agentic Backbone Distillation for Scalable Enterprise Swarm Deployment: Derives the lightweight agentic backbone from multimodal agentic trajectories in software engineering, general agents, visual design, and cybersecurity and enhances it with domain-specific reinforcement learning, enabling enterprises to deploy very accurate autonomous agent swarms at extremely low cost.

How does MiMo-V2.6 Work?

MiMo-V2.6 uses a hybrid sparse MoE backbone, unique omnimodal encoders, and a multi-stage training pipeline specifically for reinforcement learning scale that does not destabilize model representations. The backbone consists of alternating Local Sliding Window Attention (SWA) with Global Attention (GA). To ensure stable early representation learning, the first block of Transformers is designed with global attention using a dense Feed-Forward Network (FFN), while all other blocks use sparse MoE FFN without sharing any experts. Local SWA shrinks Key-Value (KV) cache memory overhead by 6× to 7×, while attention sink biases are learnable to ensure long-context coherence up to 1M tokens.

source- https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL/resolve/main/MiMo_V2_6_technical_report.pdf

Visual and audio signals are analyzed using specialized encoding streams. Vision utilizes MiMo-ViT, which is a 681M parameters' Vision Transformer with alternating row-major and column-major SWA tokenization along with spatial 2x2 merge. Audio analysis consists of two stages: AudioTokenizer (308M parameters), which includes 20 RVQ codebooks at 25Hz, and Audio Patch Encoder (127M parameters), which includes grouping every four frames to decrease token frequency to 6.25Hz. Mid-training context length has been increased from 32K → 256K → 1M. For weight optimization, AdamW method is being replaced by Muown (variant of Muon method with row-norm optimization) to ensure high data efficiency and avoid spectral norm drift and loss spikes during large batch training.

source- https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL/resolve/main/MiMo_V2_6_technical_report.pdf

The post-training procedure utilizes 'You Only RL Once' methodology by employing an asynchronous Group Relative Policy Optimization (GRPO) process on a batch of various tasks (coding, general agents, visual design, cybersecurity, context following) of size G=16 (25,000 rollouts per step). It is important to note that MoE routers remain frozen during the process of RL training. This ensures that there is no issue with router drift or expert-load collapse, which used to increase coefficient of variation from 0.78 to 2.0 and maximum expert load from 6x to 16x. The efficiency of post-RL training is improved via Multi-Prefix Multi-Teacher On-Policy Distillation (MOPD^2) using autonomous student rollouts along with prefix-conditioned single-turn rollouts. The Sample Mixer architecture manages extreme execution variances (up to 90× token length and 66× duration variance) through Adaptive Scheduling and Predictive Dispatch.

Performance Evaluation with Other Models

On the Artificial Analysis Intelligence Index v4.3, the performance score of MiMo-V2.6-Pro comes out to be 46.32, marking it as the best-performing open-source foundation model in the world. This model beats open-weights contenders like Kimi K3 and Qwen3.8 Max, and keeps up well with the top-performing proprietary frontier models like Claude Opus 5 and GPT-5.6 Sol. The main importance of the performance is in the intelligence-to-cost ratio because the model performs at the top level and offers the same cheap API cost as V2.5 generation.

source - https://mimo.xiaomi.com/mimo-v2-6

On the DeepSWE v1.1 benchmark for long-horizon software engineering agents shown in Table below, MiMo-V2.6-Pro scores 71.9, whereas MiMo-V2.6-Flash scores 67.9. This shows a massive rise of 52.9 points from the previous generation MiMo-V2.5-Pro. In this case, MiMo-V2.6-Pro beats Claude Fable 5 and stands on par with top-performing proprietary models like Claude Opus 5, GPT-5.6 Sol, and DeepSeek V4.1 Flash. The main importance of the performance is that scaling agentic reinforcement learning compute directly results in better repository-level debugging and autonomous multi-file editing.

source- https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL/resolve/main/MiMo_V2_6_technical_report.pdf

In terms of general agential workflow processes, MiMo-V2.6 excels in SaaS API orchestration through AutomationBench v1.0.6 and performs best in command line problem-solving through Terminal Bench 2.1, Model Context Protocol integration on Toolathlon-Verified, and economic wealth generation through GDPVal 2.1 among the open-weights benchmarks. In specific task fields, the model family achieves outstanding results in cybersecurity vulnerability replication through CyberGym and front-end visual coding through MiMo VisualCoding. Although there is a notable lag in performance in the field of competitive programming through ProgramBench and offensive exploit creation through ExploitGym, MiMo-V2.6 demonstrates the effectiveness of multi-task reinforcement learning in transferring tool-use skills into different agent environments.

How to access and use MiMo-V2.6?

Model weights, training logs and codebase of MiMo-V2.6 are completely open-sourced with MIT license. Developers and researchers can have access to the model checkpoints through the Hugging Face and ModelScope repositories . Models can be served using popular inference engines like vLLM or SGLang in an offline environment. For local run, hardware specifications depend on variant. Live telemetry, technical documentations and web demos can be found at the official website from Xiaomi.     

Limitations

The technical report documents several real-world infrastructure failure modes encountered during 1,000+ GPU RL training runs. These include hardware-level GPU memory double-bit errors (DBE), grader network unreachability, partial-rollout memory pool exhaustion, and expert-parallel activation Out-Of-Memory (OOM) spikes caused by up to 30x load imbalances across ranks prior to freezing routers. Host CPU OOM bottlenecks during trajectory packing also presented challenges.

Potential Architectural Advancements & Future Directions

In the pursuit of further advancing on the concept of omnimodal reinforcement learning, could future developments in this model include an entropy-aware, adaptively-controlled process for unfreezing routers post-freezing in the learning process? Through the combination of the localization of gradient values and dynamic expert load balancing, it is possible to incorporate this strategy, which would allow for the late adaptation of routers without the problems associated with experts load collapsing and activation memory spiking in large rollout clusters.

Also, in order to tackle host memory constraints in the trajectory packing process, could the adoption of a zero-copy, GPU-based memory pool in the sample dispatching pipeline be achieved, thus enabling continuous streamlining of multimodal rollouts? Lastly, could it be possible to unify visual and audio encoding processes in one cross-modal latent space?

Conclusion

From MiMo-V2.6, one can see that increasing reinforcement learning compute post-training across different verifier backed environments helps to get model capability improvements faster compared to increasing data pre-training. MiMo-V2.6 shows that it is capable of achieving frontier-level execution of tasks in software engineering, 3D creations, and tool orchestration. Xiaomi is an auditable foundation for autonomous agents.

Sources
https://mimo.xiaomi.com/mimo-v2-6
https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL/resolve/main/MiMo_V2_6_technical_report.pdf
https://huggingface.co/collections/XiaomiMiMo/mimo-v26
https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL
https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL
https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B

Disclaimer - This article is intended purely for informational purposes. It is not sponsored or endorsed by any company or organization, nor does it serve as an advertisement or promotion for any product or service. All information presented is based on publicly available resources and is subject to change. Readers are encouraged to conduct their own research and due diligence.

Jev: System One Decision Engine Delivering 193.6x Faster Execution

 

Introduction

In modern software engineering, there is a need to reconcile two very different paradigms – the deterministic predictability of compiled programs and the probabilistic nature of machine intelligence. Traditionally, implementing machine learning into a product entailed compromising heavily – wrapping the nondeterministic text streams into validation routines, dealing with the schema hallucinations, and suffering significant latency penalties. Reliable and deterministic output capable of being used by software directly thus became one of the key architectural challenges. In addition, it is necessary to have an architecture providing reliable software integration and full control of the logic, which means being able to avoid parsing errors at runtime and not break the program flow due to an invalid payload.

Implementing the decision-making models into the enterprise-scale infrastructure requires an architecture providing massive scalability in the production setting, while also becoming much faster and cheaper than multi-billion parameter autoregressive models traditionally used as baselines. Jev, an implementation of a special class of System One models by TypeSafe AI, solves this systemic issue by exposing the structured probabilistic judgment to the application code. It achieves this without relying on token-by-token generation and the need to evaluate the state in just one pass.

What is Jev?

Jev is an advanced, non-autoregressive System One decision engine developed by TypeSafe AI, the company co-founded by the former OpenAI employee Diogo Almeida. The engine was specifically designed for machine-to-machine reasoning, not for human-machine communication, and is capable of taking application state in unstructured form, including plain text, execution traces, or even JSON objects along with typed queries.

Key Features of Jev

• Non-Autoregressive Forward-Pass Evaluation: Evaluates input state and structured queries in a non-autoregressive forward pass, without any token-by-token autoregressive generation to avoid latency and execution overhead.
• Parallel Hardware-Aware Sampler: Scores multiple independent questions in parallel to a single shared context window without any linear accumulation of latency.
• Absolute 100% Type Safety: Avoids string generation and attains a zero type error rate along with eliminating the need for any schema validation code, regex parsing, and run-time retries.
• Three AI Building Blocks: Intelligence is expressed in terms of only three structured query types: Choice (picking one option from a defined list of 255 options), Score (rating state on a particular numerical or ordinal rubric) and Noul (testing boolean true/false condition with a 0 to 1 probability distribution).
• Probability Output Trained Through RLCD: Trained using RLCD instead of RLHF-based preference alignment that ensures statistically calibrated probability outputs.
• Unilateral Metering Scheme: Charged in terms of input tokens rate of $0.042 per million tokens ($42/B tokens) and with output tokens being offered absolutely free of cost ($0.00).

Use Cases of Jev

• Speculative Fan-Out State Assessment: Running multiple-variable policy assessment, routing policies, and compliance policies all at once on a single payload, so the software can use all of the decision outcomes in a single atomic operation.
• Low-Latency Command Safety & Security Firewall: Acting as a synchronous, pre-execution guard that filters raw API calls and shell commands for prompt injection attacks or policy breaches before executing them downstream.
• Reflexive Physical Control in Real-Time in Physical and Simulated Worlds: Acting as a real-time decision engine for simulated worlds, game engines, and robots who need to perform reflexive actions in response to observational data.
• Verification Layer for Generative LLMs: Acting as an additional validation layer placed after traditional conversational models that check if the responses are factually correct, comply with policies and have a proper form before sending responses to the end-user.
• Confidence-Controlled Choice of Agentic Tools: Being the decision brain of agentic tools, which will allow you to make a choice using the Choice primitive and immediately escalate a low confidence decision to the fall back procedure.
• High Throughput Routing & Scoring of Tickets and Emails: Sorting, scoring the level of urgency, and routing a high volume of tickets and emails directly into the database queue with no human involvement.

How Does Jev Work?

Structurally at its core, Jev changes the interaction between foundation models and application code, substituting probabilistic evaluation for the token generation process. In order to ask a question, application passes an unstructured input in the form of a JSON object, an execution log or any other text, together with a series of questions, described in Choice, Score, or Noul primitives in the form of an array of atomic questions. Instead of passing this information to autoregressive decoders generating text letter by letter, Jev transfers the state and the list of questions to the hardware-aware parallel sampler.

source - https://docs.typesafe.ai/introduction

The way the model is executed is closely associated with its post-training alignment technique called Reinforcement Learning for Calibrated Decisions (RLCD). As opposed to standard language models trained via Reinforcement Learning from Human Feedback (RLHF), which are optimized for maximizing human preference and hence suffer from mode dropping, overconfidence and deceptive certainty around decision boundaries, RLCD aims at optimizing the probabilities of the output of the model directly against its empirical classification outcomes. Hence, Jev produces mathematically calibrated confidence scores along with every typed value.

Potential Architectural Enhancements

Could it be possible to compile Jev’s non-autoregressive, parallel sampling engine on edge-level neural processing units (NPUs) or on specialized FPGA acceleration boards? Moving this engine from the hosted environment of cloud nodes to hardware-based environments can open up the opportunity to have less than 10 milliseconds reflex loops for embedded robots, navigation fleets, and industrial IoT. Moreover, adding state differencing along with the activation cache will enable the monitoring system to analyze live data streams without re-tokenizing static backgrounds.

On the other hand, concerning post-training and research side, extension of the RLCD framework by cross-modal joint embeddings will enable direct probability-based evaluation of video streams, audio data, spatial sensors without any text transduction at all. To overcome one-pass cardinality limitations, it is possible to implement dynamic and speculative tree cascading right into the forward pass. Such modification will enable hierarchical routing of decisions across thousands of available options in one-pass manner and still guarantee type safety.

Performance Evaluation with Other Models

For measuring the efficiency of decision-making in real-world scenarios, TypeSafe AI performed the performance evaluation of other models with the help of structured compute graphs of the real-life production business logic and not through any static benchmark datasets. The baseline reference distributions have been derived based on the average results of the leading frontier models such as GPT-6 Astra and Claude Fable 5.1. In order to perform comparative tests, all the standard LLMs were evaluated using TypeSafe AI’s System One, which forced them to output only structured decisions and probabilities.

source - https://typesafe.ai/blog/introducing-system-one-models-and-jev

Jev showed an extremely radical difference when tested with the speed and latency benchmarks. Where the standard frontier LLMs needed 3 to 329 seconds to execute structured decision workflows, Jev completed the task within 70ms to 500ms. On representative production workflow benchmarks, the time required to perform decision cycles was 0.114 seconds for Jev against 8.566 seconds for the standard LLMs, which gave a 193.6x improvement. Moreover, per-workflow execution costs were reduced from $0.013880 for standard frontier models to $0.000081 for Jev, giving a cost reduction of 444.6x.

The synopsis of the benchmarks highlights a paradigm shift in production economics and reliability of systems. The fact that Jev achieves the feat of having a type error rate of 0% means it removes validation delay, regex parsing, and retries that are a constant feature of conventional LLMs. Input pricing comes down to $0.042 million tokens, which is 238x cheaper than the Claude Fable 5.1. Output tokens come free of charge, making Jev bring back the Pareto frontier for workflow intelligence to cost ratio.

How to Access and Use Jev?

The proprietary and hosted Jev API is available as early access on the TypeSafe AI platform through an onboarding waitlist. The model cannot be executed locally as there are no open-weights or self-hosted versions of the model available yet. The integration of Jev into production-ready systems can be done using the official Client SDKs that have been provided in Python and JavaScript/TypeScript or directly via REST API.

Limitations and Future Work

Despite the efficiency that Jev is able to offer in terms of speed and reduced costs, the unique architecture used in Jev has limitations of function as well. First, it does not support open-ended string creation at all making it unusable in conversational chatbots, copilots and generative code creation. Built solely for single pass System 1 classification, it cannot perform multi-step inference or analysis of multi-faceted tradeoffs in one prompt and requires developers to break down problems into smaller atomic questions within application logic. The Choice primitive supports a maximum cardinality of 255 choices – thus, requiring a slower process of scoring in two stages when dealing with greater number of choices – as well as limited input states which include only text and data.

In order to overcome these limitations, the technical road map of TypeSafe AI includes development in areas of expanding input types to cover visual and multimodal data, System One model pipeline extension, developer workshops and hackathons and onboarding from the waitlist. Future work will focus on showing computational and economic viability of its parallel sampling architecture and RLCD training.

Conclusion

Jev demonstrates that specialization of foundation models for non-autoregressive decision execution provides outstanding benefits. It even demonstrates that separation of probabilistic reasoning and text generation allows for creating a fast, affordable, and type-safe primitive that becomes a part of the deterministic codebase. For infrastructure professionals looking for the ways to automate the logic without significant latency and validation costs, Jev sets a benchmark for the future of machine-native AI infrastructure.

Sources
https://typesafe.ai/blog/introducing-system-one-models-and-jev
https://typesafe.ai/
https://docs.typesafe.ai/introduction

Disclaimer - This article is intended purely for informational purposes. It is not sponsored or endorsed by any company or organization, nor does it serve as an advertisement or promotion for any product or service. All information presented is based on publicly available resources and is subject to change. Readers are encouraged to conduct their own research and due diligence.

Nex-N2.5: Open-Weight AI That Controls Web Browsers and Desktop OS

Introduction

There is currently an enormous paradigm shift taking place in modern sparse open-weight architectures in how they process multimodal streams. Rather than treating graphical streams as passive data input or using text-based command chains only, today’s architectures such as Nex-N2.5 take advantage of optical perception streams for ongoing runtime error correction. This architectural advance provides a way of overcoming the frailties of traditional automation pipeline design, which cannot cope with minor interface rendering errors or unexpected terminal outputs during multi-turn processes.

It is important to strike a balance between selective parameter passing and efficiency when running autonomous extended processes in the enterprise environment. Being able to adapt the computing infrastructure for advanced computation to work for workflow is key to achieving efficient execution in software engineering, web navigation, and interface control. It is through making such hardware optimization work in combination with closed-loop image validation to provide self-rectification in an autonomous manner that modern architectures such as Nex-N2.5 achieve desktop agency and multi-step reasoning without dense latency cost.

What is Nex-N2.5?

Nex-N2.5 is a family of Mixture-of-Experts (MoE) models developed by Nex-AGI, explicitly purpose-built for visually grounded agency, native desktop and web GUI navigation, autonomous software development, and closed-loop execution. Transitioning from traditional text-heavy agent frameworks, the model utilizes continuous visual perception as a real-time verification interface to execute code, manage operating systems, navigate complex web applications, and self-correct system actions based on observed environmental states.

Model Variants

• Nex-N2.5-mini: Runs on 35 Billion total parameters (MoE) with 3 Billion active parameters per token (A3B). Built on top of the Qwen3.5-35B-A3B-Base multimodal foundation, this model is designed to achieve fast instruction-following, low-latency API serving, and on-the-fly tool execution. Efficiently running on a single node featuring 2×H100 GPUs with Tensor Parallelism (TP=2).
• Nex-N2.5-Pro: Equipped with 397 Billion total parameters (MoE) with 17 Billion active parameters per token (A17B) and built on top of the Qwen3.5-397B-A17B multimodal base, this is the main workhorse model for complex reasoning, multi-agent orchestration, and all-in-one developer stack. Designed for efficient running on a single node with 8×H100 GPUs (TP=8) using the dedicated NexRT inference engine.
• Nex-N2.5-Max: This is the flagship variant equipped with 1.6 Trillion total parameters (MoE) with 49 Billion active parameters per token (A49B). In contrast to other variants, this one uses the DeepSeek-V4-Pro-Base text-only foundation model for deep reasoning, scientific research modeling, and high-level architectural code generation. Efficiently running on multi-node clusters.

Key Features of Nex-N2.5

• Visually Grounded Agency & Closed Loop Self Correction: Controls web browsers and desktop operating systems through a visually grounded feedback loop. This model performs visual action execution, visual UI update evaluation, detects rendering bugs/page anomalies, and dynamically re-plans execution in real time, without any human involvement.
• Native GUI Navigation & Normalized Grounding: Can perform exact cursor-mouse actions in both desktop and web applications. For achieving spatial precision in different resolutions (OSWorld, WebArena, and WebTest benchmarks), normalized grounding is used in a standardized 0–1000 spatial grid.
• Coherent Logic & Multi-Step State Traversal: Maintains coherent reasoning capabilities across the entire process of task decomposition, strategic changes, visual state traversal, and self assessment within complex context switches.
• Trillion Parameter Scale Post Training Pipeline: Represents the first time Nex-AGI is attempting a post training process at a 1.6 trillion parameter scale. The pipeline receives live terminal output, web DOM tree, and screenshot streams directly into the training loop.
• Heterogeneous Base Foundation Models: Uses a combination approach by utilizing multimodal Qwen3.5 models with visual spatial interactions (Mini and Pro) as well as a 1.6T text only DeepSeek-V4-Pro base (Max) for logical synthesis.

Use Cases of Nex-N2.5

• Dual Model Autonomous Code Refactoring & Visual Playtesting: Facilitates automation of CI/CD pipelines where the state-of-the-art reasoning model rewrites complex codebase, and a multimodal model serves as the workhorse that builds up the software, opens the GUI, validates the UI rendering against design specifications and sends the screenshot differences to the code refactoring model to perform corrections before merging.
• Closed Loop Visual GUI Navigation with Autonomous Corrections: Navigates through browser and desktop applications checking the visual state after each click, and autonomously changes the path execution if unexpected popups appear or sites are broken.
• Multi-Backbone Models Inference under a Single Gateway: Simplifies enterprise inference management through serving of lightweight, medium, and ultra-heavy models under one standardized API gateway with custom server-side reasoning and tool-parsing routers.
• Zero-Script Legacy ERP and Desktop Workflow Automation: Automates multi-step workflows within enterprise applications including legacy ERP and desktop applications as well as mainframes emulators, using screen vision and normal coordinates control, without the use of unreliable APIs and RPA scripts.
• Engineering Task Criticality-Based SLA-Allocation of Compute Resources: Minimizes company compute cost by allocating developer traffic depending on the criticality of their tasks, sending code completion to high performance 2× H100 machines, feature requests to mid-performance 8× H100 machines and architecture-related tasks to high end multi-node machines.
• Cost Margin and Latency Management via Reasoning Control: Controls product margins and latencies by configuring API requests to skip the thinking trace, use adaptive thinking trace or deep reasoning trace depending on the criticality of the request.

How Does Nex-N2.5 Work?

Nex-N2.5 integrates an execution-feedback loop into its post-training flow, end-to-end. Instead of making predictions for the next tokens based only on pairs of instruction and response that don’t change during training, the training procedure makes the model experience the live execution environment by receiving terminal output streams, web DOM tree structures, and desktop screenshot streams in real-time. Consequently, the model learns to map raw vision observations directly to useful tool invocations and normalized keyboard/mouse coordinates on a 0-1000 spatial grid.

To enable efficient decoding in real-time on the 8xH100/H200 GPU cluster of Nex-N2.5-Pro, Nex-AGI has developed the dedicated inference engine NexRT. The engine supports Standard Decoding with Multi-Token Prediction (MTP) and DFlash block-diffusion. The engine features custom CUDA kernels, full decode CUDA Graphs, GPU Direct between nodes communication, and Context-Parallel Attention (CP) that allows avoiding duplicate KV cache reads with extended context windows.

Performance Comparison with other Models

In the context of web browsing and long context information synthesis, BrowseComp draws attention to Nex-N2.5-Max which has achieved the #1 rank worldwide and is directly better than some of the best proprietary frontier models like Claude Opus 5 and GPT-5.6 Sol. In addition to that, under the same evaluation paradigm, Pro and mini versions have shown significant generational improvements compared to their predecessors. While showing superiority in web browsing, Nex-N2.5-Pro has achieved the #1 rank worldwide in OSWorld-G spatial grounding benchmarks, beating other best visual and multimodal models like Qwen3.8-Max, GLM-5.3-Flash, GPT-5.6 Sol, and Claude Opus 5.

source - https://nex-agi.com/

When it comes to software engineering for repositories and debugging multiple files, the evaluations on SWE-Bench Pro and DeepSWE v1.1 show how well the model can generate code and solve problems. The model, compared to leading open-weight models such as DeepSeek-V4-Pro and GLM-5.3, shows better results on most of the important agentic, web, and programming benchmarks, while on other benchmarks, such as SWE-Bench Pro and GUI navigation in space, it shows competitive performance with some targeted leads. This good performance showcases the efficiency of the model suite in dealing with large full-stack codebases and autonomous engineering problems.

source - https://nex-agi.com/

In all types of agentic workflows, desktop operating system controls, and multi-turn execution of tools, Nex-N2.5 remains consistently powerful in different evaluation sets. In all evaluation sets for workflow and knowledge performance such as GDPval-AA v2, AutomationBench, and Toolathlon Verified, the Max version is superior to top frontier agents such as GPT-5.6 Sol, GLM-5.3, Kimi-K3, and DeepSeek-V4-Pro. Also, the excellent performance on Terminal-Bench, OSWorld-Verified, and OSWorld-2 shows the cross-domain versatility of the family of models, proving that the feedback loop for perception and execution in reality is successful.

How to Access and Use Nex-N2.5?

Model Weights for Nex-N2.5 are available under the Apache-2.0 License through Hugging Face and ModelScope, while the codebase is available at GitHub. Hosted API endpoints can be accessed through OpenRouter.  Reasoning efforts during API calls can be dynamically controlled through 'reasoning_effort'. During SGLang deployment, launching scripts require providing '--tool-call-parser qwen3_coder', as well as corresponding reasoning parsers: '--reasoning-parser qwen3' for Mini/Pro or '--reasoning-parser deepseek-r1' for Max.

Limitations and Future Work

Using the Nex-N2.5 framework involves significant overhead in terms of hardware infrastructure as the Max model needs at least 16× H200 GPUs in 2 nodes with DeepEP/DeepGEMM networking, whereas the Pro model needs an 8× H100 node with SGLang patches for ideal token decoding. Going forward, work will focus on using trillion-scale post-training knowledge for larger foundation backbones and complete open-source availability of the NexCUA evaluation framework.

Conclusion

Nex-N2.5 presents an implementation strategy for open weight agentic architectures in transforming computer vision into a proactive visual execution cycle instead of a passive description method. It illustrates the way in which enterprise automation systems can move away from fragile brittle scripting to visual self-correcting autonomy through the use of precise spatial grounding along with customized inference engines such as NexRT.

Sources:
https://nex-agi.com/
https://github.com/nex-agi/Nex-N2.5
https://huggingface.co/collections/nex-agi/nex-n25
https://huggingface.co/nex-agi/Nex-N2.5-Max
https://huggingface.co/nex-agi/Nex-N2.5-Pro
https://huggingface.co/nex-agi/Nex-N2.5-mini


Disclaimer - This article is intended purely for informational purposes. It is not sponsored or endorsed by any company or organization, nor does it serve as an advertisement or promotion for any product or service. All information presented is based on publicly available resources and is subject to change. Readers are encouraged to conduct their own research and due diligence.