# MAI Thinking 1 ![Organization](https://img.shields.io/badge/Org-Microsoft-0078d4) ![License](https://img.shields.io/badge/License-Proprietary-d29922) ![Released](https://img.shields.io/badge/Released-May%202025-555) ![Reasoning](https://img.shields.io/badge/Type-Reasoning-8250df) > **Quick answer:** MAI Thinking 1 is Microsoft's May 2025 proprietary reasoning model, scoring 85% MMLU-Pro, 73.5% SWE-Bench Verified, and 72.0% BFCL v3. It represents Microsoft's in-house reasoning model capability, distinct from their Azure-hosted OpenAI models. ## At a Glance **Where MAI Thinking 1 leads** - 85% MMLU-Pro — strong academic knowledge with reasoning - 73.5% SWE-Bench Verified — competitive software engineering - 72.0% BFCL v3 — good function calling with reasoning - Microsoft-built: not an OpenAI model resold **Where it lags** - Proprietary: no open weights - February 2025 knowledge cutoff — older than competing models - Limited public benchmark data - Text-only **Best for:** Azure-native deployments requiring Microsoft-built reasoning models; teams needing MAI capabilities in Azure AI Foundry. ## What MAI Thinking 1 Is MAI Thinking 1 (released May 2025) is Microsoft's first publicly released in-house reasoning model — distinct from the GPT models that Microsoft distributes via Azure OpenAI. "MAI" stands for Microsoft AI, and the "Thinking" designation indicates a chain-of-thought reasoning architecture. The model is positioned as Microsoft's answer to OpenAI's o-series for teams that prefer Microsoft-built models over OpenAI-built ones in Azure. With 85% MMLU-Pro and 73.5% SWE-Bench, it is competitive with models at a similar capability tier from the same period. ## Specifications | Field | Value | |---|---| | Organization | Microsoft | | License | Proprietary (API only) | | Release date | May 2025 | | Knowledge cutoff | February 2025 | | Modality | Text only | | Architecture | Chain-of-thought reasoning | ## Pricing Available via Microsoft Azure AI Foundry. Refer to Azure AI pricing for current MAI Thinking 1 rates. ## Public Benchmark Scores | Benchmark | Score | Source | Date | |---|---|---|---| | [MMLU-Pro](/benchmarks/mmlu-pro) | 85% | Benchgen evaluation | 2025-07 | | [SWE-Bench Verified](/benchmarks/swe-bench-verified) | 73.5% | Benchgen evaluation | 2025-07 | | [BFCL v3](/benchmarks/bfcl-v3) | 72.0% | Benchgen evaluation | 2025-07 | ## MAI Thinking 1 vs Alternatives | Model | MMLU-Pro | SWE-Bench | BFCL v3 | Provider | |---|---|---|---|---| | **MAI Thinking 1** | 85% | 73.5% | 72.0% | Microsoft | | [Phi-4 Reasoning Plus](/models/microsoft/phi-4-reasoning-plus) | 76% | — | — | Microsoft | | [GPT-5 Codex](/models/openai/gpt-5-codex) | — | 74.5% | — | OpenAI | | [o3-mini](/models/openai/o3-mini) | — | — | — | OpenAI | MAI Thinking 1 vs Phi-4 Reasoning Plus: higher MMLU-Pro (85% vs 76%) at likely higher cost (proprietary API vs open weights). MAI Thinking 1 is for teams requiring Microsoft-built proprietary reasoning vs open-weight options. ## Related Models & Benchmarks - Compare: [Phi-4 Reasoning Plus](/models/microsoft/phi-4-reasoning-plus) · [GPT-5 Codex](/models/openai/gpt-5-codex) · [o3-mini](/models/openai/o3-mini) - Benchmarks: [MMLU-Pro](/benchmarks/mmlu-pro) · [SWE-Bench Verified](/benchmarks/swe-bench-verified) · [BFCL v3](/benchmarks/bfcl-v3) ## Frequently Asked Questions

What is MAI Thinking 1?

Microsoft's May 2025 in-house reasoning model scoring 85% MMLU-Pro, 73.5% SWE-Bench Verified, and 72.0% BFCL v3. Microsoft-built, distinct from Azure OpenAI's GPT models.

Is MAI Thinking 1 the same as an OpenAI model?

No — MAI (Microsoft AI) is Microsoft's in-house model development. Azure OpenAI models (GPT-4o, o3, etc.) are built by OpenAI. MAI Thinking 1 is built by Microsoft's own AI research team.

--- *Specs from Microsoft's MAI Thinking 1 release (May 2025) and Benchgen evaluations. Last updated 2026-07-24.*

MAI-Thinking-1 — Evaluations

Evaluations