Microsoft releases MAI-Thinking-1 reasoning model at 35B parameters, MAI-Code-1-Flash for GitHub Copilot
Microsoft announced two new language models: MAI-Thinking-1, a 35B parameter reasoning model available to select early partners, and MAI-Code-1-Flash, a 5B parameter coding model rolling out to GitHub Copilot individual users in VS Code. Both models were trained on commercially licensed data without distillation from third-party models.
Microsoft releases MAI-Thinking-1 reasoning model at 35B parameters, MAI-Code-1-Flash for GitHub Copilot
Microsoft announced two new language models on June 2nd, 2026: MAI-Thinking-1, a 35B parameter reasoning model, and MAI-Code-1-Flash, a 5B parameter coding model designed specifically for GitHub Copilot and VS Code.
MAI-Thinking-1: 35B parameter reasoning model
MAI-Thinking-1 is currently available only to select early partners. According to Microsoft, the model "is preferred to Sonnet 4.6 in our blind human side-by-side evaluations."
Microsoft claims the model was "trained from the ground up on enterprise grade, clean and commercially licensed data, without distillation from third-party models." Pricing and broader availability have not yet been disclosed.
MAI-Code-1-Flash: 5B parameters for coding
MAI-Code-1-Flash is described as "purpose-built for GitHub Copilot and VS Code to deliver high performance and lower cost." The model is rolling out to GitHub Copilot individual users in Visual Studio Code.
Microsoft states the model is "built end-to-end by Microsoft using clean and appropriately licensed data." Specific details about the training data sources have not been provided.
Technical specifications
- MAI-Thinking-1: 35B parameters, reasoning-focused
- MAI-Code-1-Flash: 5B parameters, code-specialized
- Training data: Both models claim to use commercially licensed data without third-party model distillation
- Availability: MAI-Thinking-1 limited to early partners; MAI-Code-1-Flash rolling out to GitHub Copilot users
Context window size, benchmark scores on standard evaluations like HumanEval or MBPP, and pricing details have not been disclosed.
What this means
Microsoft's focus on small parameter counts—35B and 5B—runs counter to the industry trend toward larger models. If MAI-Thinking-1's claimed performance against Claude Sonnet 4.6 holds up in independent testing, it would represent a significant efficiency gain for reasoning tasks.
The emphasis on "appropriately licensed" training data is notable given ongoing legal challenges around AI training datasets. If verified, these could be among the first commercially viable code models trained without scraping unlicensed code repositories, though Microsoft has not provided specifics on what "appropriately licensed" means in practice.
Related Articles
Microsoft's ThinkingBox-Training lifts Qwen3.8-27B fully reliable task completions from 38 to 107 of 507
Microsoft and Hugging Face released ThinkingBox-Training, which turns executable database-state checks into RL rewards. After 50 GRPO steps, Qwen3.8-27B's pass@1 rose from 51.70% to 60.78%, and tasks passed on all 20 attempts rose from 38 to 107, according to the authors.
Microsoft's MAI-DxO scores 83% vs 36% for physicians on complex NEJM cases in Science study
Microsoft's MAI-DxO, a multi-agent diagnostic orchestrator, was correct on 83% of complex New England Journal of Medicine cases in simulation, versus 36% for generalist physicians, according to a study in Science. The data suggests most of the gain comes from stronger base models, with orchestration adding a smaller lift, mainly on complex cases.
MOSI and OpenMOSS open-source MOSS-Image, a 5B text-to-image model for dense text and JSON layout control
MOSI Intelligence and OpenMOSS have open-sourced MOSS-Image, a 5B-parameter text-to-image model focused on dense text and JSON layout control. According to Pandaily, it quickly topped Hugging Face's text-to-image trending list.
Gradium open-sources Phonon 31m, a 31M-parameter on-device TTS model under Apache 2.0
Gradium has open-sourced Phonon, a family of on-device text-to-speech models. The English Phonon 31m has 31M parameters, ships under Apache 2.0, fits in 64.5 MB, and starts speaking in about 7 ms on a laptop CPU.
Comments
Loading...