Skip to main content

    Microsoft

    MAI-Thinking-1

    Microsoft's first fully in-house reasoning model, unveiled at Build 2026 (June 2) — a sparse MoE with ~35B active of ~1T total params, trained without OpenAI distillation using data Microsoft describes as appropriately licensed. AIME 2025 97.0%, AIME 2026 94.5%, SWE-Bench Pro competitive with Claude Opus 4.6, and preferred over Sonnet 4.6 in blind human evals across 1,276 tasks. 256K context; public preview on Microsoft Foundry since August 12, 2026 with third-party availability via Fireworks AI, Baseten, and OpenRouter.

    256K tokens
    Text only
    Max
    Provider
    Microsoft
    Context
    256K
    Estimated · As of Aug 11, 2026Stale · 31d old
    yno subscription tier
    Max
    Released
    Jun 2, 2026
    Speed
    Medium
    Reasoning
    Expert
    Modality
    Text only

    Benchmarks

    Quality

    AA Intelligence Index
    46points
    Estimated · As of Aug 11, 2026Stale · 31d old
    GPQA Diamond
    93%
    Estimated · As of Aug 11, 2026Stale · 31d old

    Speed and latency

    Output speed
    95tokens/s
    Estimated · As of Aug 11, 2026Stale · 31d old

    API pricing

    Not available

    AA measurements unavailable

    Indexes use points; evaluations use accuracy percentages. API measurements do not measure yno application performance. Methodology

    Capabilities

    • Reasoning
    • Mathematics
    • Code
    • Function Calling
    • Chain of Thought

    Use cases

    • Mathematical reasoning
    • Software engineering
    • Azure/Foundry deployments
    • OpenAI-independent Microsoft stack

    Strengths

    • AIME 2025 97.0%
    • No OpenAI distillation
    • ~35B active of ~1T MoE
    • Beat Sonnet 4.6 in blind evals

    Best for

    Microsoft-stack teams that want a first-party frontier reasoning model

    Use MAI-Thinking-1 in yno.ai

    No credit card required

    Related models

    Frequently asked

    What is MAI-Thinking-1?
    Microsoft's first fully in-house reasoning model, unveiled at Build 2026 (June 2) — a sparse MoE with ~35B active of ~1T total params, trained without OpenAI distillation using data Microsoft describes as appropriately licensed. AIME 2025 97.0%, AIME 2026 94.5%, SWE-Bench Pro competitive with Claude Opus 4.6, and preferred over Sonnet 4.6 in blind human evals across 1,276 tasks. 256K context; public preview on Microsoft Foundry since August 12, 2026 with third-party availability via Fireworks AI, Baseten, and OpenRouter.
    How much does MAI-Thinking-1 cost in yno.ai?
    MAI-Thinking-1 is available on the Max tier of yno.ai.
    What can MAI-Thinking-1 do?
    MAI-Thinking-1 is best for Microsoft-stack teams that want a first-party frontier reasoning model. Its main capabilities include Reasoning, Mathematics, Code, Function Calling, Chain of Thought.
    How does MAI-Thinking-1 compare to other models?
    MAI-Thinking-1 excels at AIME 2025 97.0% and is recommended for Mathematical reasoning, Software engineering, Azure/Foundry deployments, OpenAI-independent Microsoft stack. See related models below.
    How do I use MAI-Thinking-1 in yno.ai?
    Sign up for yno.ai, select MAI-Thinking-1 from the model picker, and start chatting.