MAI-Thinking-1: benchmark results
Microsoft AI's first in-house reasoning model, a sparse MoE (1T total, 35B active). Provider: Microsoft. Released 2026-06-02. Access: API.
Aggregate Rating 1700 ± 24, rank #346 of 10916 rated models, from 11 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Arcee MMLU-Pro (BenchLM) | 85 | Accuracy (%) | 73.6 |
| ZeroEval GPQA Diamond | 84.2 | GPQA Diamond Score | 70.2 |
| LLM Stats Score | 32.4 | LLM Stats Score (conservative rating) | 68.1 |
| LLM Stats (AIME 2026) | 94.5 | Score (%) | 63 |
| LLM Stats (LongBench v2) | 61 | Score (%) | 61.1 |
| BenchLM | 52.8 | Overall Score | 55.8 |
| LLM Stats (SimpleQA Verified) | 31 | Score (%) | 50 |
| MedXpertQA | 43 | Score (mixed sources) | 46.7 |
| LLM Stats (Multi-Challenge) | 53 | Score (%) | 40.6 |
| LLM Stats (HMMT Feb 26) | 84.9 | Score (%) | 33.3 |
| LLM Stats (HealthBench Professional) | 35 | Score (%) | 0 |
Interactive version: theaggregate.ai/model?slug=mai-thinking-1 · How It Works · Data refreshed daily, snapshot 2026-11-08.