MAI-Thinking-1
Microsoft · released
Microsoft AI's first in-house reasoning model, trained from scratch for maths, coding and enterprise agent work. It is text-only and sold through Microsoft Foundry — and none of the independent evaluators tracked here has published results for it yet.
Best for: Azure-native workloads · Coding and maths
- Intelligence Index
- —
- Not yet evaluated
- Price per 1M tokens
- $2 in $8 out
- Blended $3.5 at 3:1
- Context window
- 256K
- Up to 64K output
- Output speed
- —
- Not measured
Capability profile
One 0–100 score per area, placing this model between the weakest and strongest tracked model across that area's benchmarks.
- OverallNot enough data
Based on 0 of 3 benchmarks
- CodingNot enough data
Based on 0 of 5 benchmarks
- Agents & knowledge workNot enough data
Based on 0 of 6 benchmarks
- Reasoning & scienceNot enough data
Based on 0 of 2 benchmarks
- Knowledge & reliabilityNot enough data
Based on 0 of 2 benchmarks
- VisionNot enough data
Based on 0 of 2 benchmarks
Benchmark results
All third-party results on record. Rank is among the tracked models that have been tested on that benchmark.
Overall
- Artificial Analysis Intelligence IndexNot tested
- Vals IndexNot tested
- Arena · TextNot tested
Coding
- Terminal-Bench 4.0Not tested
- SciCodeNot tested
- Arena · WebDevNot tested
- Vibe Code BenchNot tested
- Code MigrationNot tested
Agents & knowledge work
- GDPval-AANot tested
- Arena · AgentNot tested
- Finance Agent v2Not tested
- Excel Modeling BenchmarkNot tested
- Legal Research BenchNot tested
- Tax Agent BenchNot tested
Reasoning & science
- Humanity's Last ExamNot tested
- CritPtNot tested
- ARC-AGI-2saturatingNot tested
- ARC-AGI-3Not tested
- GPQA DiamondsaturatingNot tested
Knowledge & reliability
- AA-OmniscienceNot tested
- AA-LCRNot tested
Vision
- MMMU-ProNot tested
- Arena · VisionNot tested
Reported by Microsoft
Numbers the lab published at launch, run on its own harness and settings. Useful context, but not independently verified and never used in the rankings on this site.
- SWE-bench Pro · average of 4 runs52.8%
- SWE-bench Verified · average of 4 runs73.5%
- GPQA Diamond · average of 4 runs84.2%
Specifications
- Lab
- Microsoft
- Released
- Jun 2, 2026
- Availability
- Preview
- Context window
- 256,000 tokens
- Max output
- 64,000 tokens
- Accepts
- Text
- Knowledge cutoff
- Jul 2025
- Weights
- Closed
- Parameters
- ~1T total · 35B active
- Reasoning control
- Always on
- API model ID
mai-thinking-1
Announced on 2 June in private preview; in public preview on Microsoft Foundry since 12 August.
Pricing
- Input / 1M tokens
- $2
- Cached input / 1M
- $0.20
- Output / 1M tokens
- $8
- Blended (3:1)
- $3.5
Pay-as-you-go price on Microsoft Foundry.
Worth knowing
- No independent benchmark results were available at this snapshot, so it is listed but not ranked.
Official sources
23 benchmarks tracked. See methodology and sources.