Microsoft makes Decision-1.1 generally available, adds MAI-Code-2 and MAI-Image-2.7 models
Microsoft announced that Decision-1.1, a decision model built on Google's Gemma 4 26B A4, is generally available. The company also introduced MAI-Code-2, MAI-Image-2.7 and MAI-Cyber-2-Flash, plus Foundry updates for agent evaluation and continuous improvement.
Microsoft today made Decision-1.1 generally available, a model it describes as fast and cost-effective for making reliable decisions in software at scale. The company also announced new models in its MAI first-party family and updates to Foundry, its enterprise platform for grounding, testing and deploying AI applications and autonomous agents.
Decision-1.1
Decision-1.1 builds on Microsoft-Decision-1, released in October. According to Microsoft, the new model delivers a significant improvement in intelligence and speed at a competitive price. Pricing figures were not disclosed.
Microsoft says Decision-1.1 is built on a new base, Google LLC's Gemma 4 26B A4, and adds multimodal support so it can analyze images in instructions.
The pitch is aimed at tasks such as routing, classification, prioritization, verification and workflow control. Microsoft says LLMs often spend time "thinking it through" on these decisions, which it calls unnecessary. Decision models are designed to reach answers quickly within constrained definitions and to return calibrated confidence scores, so applications can defer when uncertainty is high. Microsoft claims the model offers superior decision-making compared with large language models and can decide dynamically at low cost instead of relying on hard-coded rules.
New MAI models
Microsoft extended MAI, its first-party model family in Foundry, which it positions as offering enterprise trust.
- MAI-Code-2: handles demanding agentic coding work.
- MAI-Code-2-Flash: supports high-volume, everyday coding, letting developers choose between performance and cost. Both Code-2 models are generally available on GitHub, and Code-2-Flash is in private preview in Foundry.
- MAI-Image-2.7 and MAI-Image-2.7-Flash: in preview, with what Microsoft describes as sharper editing and more consistent image generation. Microsoft says this helps brands keep a consistent look across campaigns at lower cost.
- MAI-Cyber-2-Flash: a purpose-built cybersecurity model, in preview within Codename MDASH, Microsoft's agentic code security system, which is itself in preview for eligible organizations.
Context windows, parameter counts, benchmark scores and pricing for these models were not disclosed.
Foundry updates
Foundry IQ, part of Microsoft IQ, now provides an enterprise knowledge graph out of the box, giving applications context from unstructured and application data across an organization's estate.
Microsoft also said Foundry now uses its existing transparency and observability capabilities to create learning loops. Internal evaluations let autonomous agents set baselines for quality, safety, latency and cost. Human annotations, now in preview, let experts rate and comment on agent traces based on business judgment, and that feedback enters the improvement loop. It is combined with optimization tests against evaluations and comparisons with real-world results.
Foundry also gains capabilities to adapt supported models with organizational data using reinforcement learning, with Foundry running the compute. Each improvement becomes the next baseline, so teams can keep measuring and testing changes.
On model access, Foundry now has an expanded pool featuring the top 50 models, supporting a model router and AI Gateway for model selection. It also offers more than 11,000 models from OpenAI Group PBC, Anthropic PBC, Meta Platforms Inc., Mistral AI SAS, the open-source community and others. Microsoft says supported models can be evaluated and deployed in Foundry using the same continuous improvement workflow as agents.
What this means
Decision-1.1 targets a specific pattern: agent and application pipelines that use a general-purpose LLM for routine control decisions. Microsoft's argument is that a smaller model with calibrated confidence scores can handle those steps faster and at lower cost, and can signal when a case should be deferred. Those claims come from Microsoft and no benchmark or pricing data accompanied them.
The MAI releases pair a full model with a Flash variant in coding and image generation, giving developers an explicit performance-versus-cost choice within one family. Availability is uneven: the Code-2 models are generally available on GitHub, while Code-2-Flash in Foundry, the image models and MAI-Cyber-2-Flash remain in private preview or preview.
The Foundry changes tie evaluation, human annotation and reinforcement learning on organizational data into one loop, with baselines for quality, safety, latency and cost reset after each improvement.
Related Articles
GitHub Copilot team customizations now generally available for organization teams
GitHub announced that team customizations for GitHub Copilot are generally available for organization teams. The feature lets teams share the Copilot plugins they rely on without asking every member to repeat the same setup.
Microsoft releases Aion 1.0 Instruct, a 2.3B on-device SLM with 32K context under MIT license
Microsoft's Windows Applied Sciences Group has published Aion 1.0 Instruct, a roughly 2.29B-parameter instruction-tuned small language model with a 32,768-token context window under the MIT license. A further optimized version is slated to become the Windows OS SLM by the end of January 2027.
Microsoft launches Autopilot agent, Home interface and Copilotforce with Salesforce at Ignite 2026
At Ignite 2026, Microsoft introduced Autopilot, an always-on Copilot agent with its own Entra identity, available now to Frontier program subscribers. It also unveiled a new Home interface, Copilot Workspaces, Copilot Code and Copilotforce, a Salesforce integration with 38 prebuilt skills.
Microsoft makes Word, Excel and PowerPoint in Copilot generally available, calls Copilot "the new OS for work"
Microsoft CEO Satya Nadella announced Tuesday at Ignite the general availability of Word, Excel and PowerPoint inside Copilot, which he called "the new OS for work." The company also introduced Copilot Autopilot, Microsoft IQ and expanded Agent 365 controls, with pricing for Autopilot still unannounced.
Comments
Loading...