Inference
EN
Microsoft

MAI-Thinking-1

MAI-Thinking-1 is a reasoning model from Microsoft in the MAI-Thinking family. 10 benchmarks count toward its index, in 5 categories.

  • Closed weights
  • Reasoning
  • Input: text
  • Released 2026-06
RankedEnough results for an index.
IndexAll categories, out of 100.53.5±8.1Median 51.0
CoverageIndex weight that has results.70%
SpeedOutput tokens per second.—Median 67/s
First tokenSeconds to the first output token.—Median 1.3 s
Input / 1MUS dollars per 1M input tokens.—Median $0.60
Output / 1MUS dollars per 1M output tokens.—Median $2.55
ContextMaximum tokens in one request.256KMedian 262K
EloLMArena rating and rank.—Median 1422

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.1 counted46.6—
CodingCode writing and repair.3 counted55.294 / 209
ReasoningLogic problems and puzzles.None counted——
MultimodalTasks with images and text.None counted——
KnowledgeFacts and expert knowledge.2 counted53.0112 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.1 counted68.5—
MathMath problems.3 counted52.675 / 168

Results

10 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
Instruction Following BenchmarkInstruction85.0%2 / 57
American Invitational Mathematics Examination 2025Math97.0%4 / 53
LiveCodeBench v6Coding87.7%15 / 59
Massive Multitask Language Understanding ProfessionalKnowledge85.0%20 / 75
Graduate-Level Google-Proof Q&AKnowledge84.2%55 / 141
Software Engineering Benchmark VerifiedCoding73.5%52 / 114
AIME 2026Math94.5%19 / 34
Harvard-MIT Mathematics Tournament February 2026Math84.9%17 / 27
GPQA Diamond · BenchLMKnowledge84.2%60 / 84
SWE-bench ProCoding52.8%62 / 83
Terminal-Bench 2.0 · BenchLMAgentic46.0%34 / 41
Graphwalks BFS 0K-128KReasoning90.0%—

Sources

BenchLMCC BY-NC 4.0

History

5 releases
Index54.5 → 53.5
23 Oct 202612 Nov 2026

Questions

How good is MAI-Thinking-1?
Its index is 53.5 out of 100. That is place 113 of 283 ranked models.
What is MAI-Thinking-1 best at?
Knowledge: 53.0 out of 100, place 112 of 268.
What is the context window of MAI-Thinking-1?
256K tokens.
Is MAI-Thinking-1 open source?
No. Its weights are not published.
When was MAI-Thinking-1 released?
June 2026.

Badge

MAI-Thinking-1 on Inference360: place 113 of 283, index 53.5
HTML
<a href="https://inference360.ai/models/mai-thinking-1"><img src="https://inference360.ai/badges/mai-thinking-1.svg" alt="MAI-Thinking-1 on Inference360: place 113 of 283, index 53.5" height="20"></a>
Markdown
[![MAI-Thinking-1 on Inference360: place 113 of 283, index 53.5](https://inference360.ai/badges/mai-thinking-1.svg)](https://inference360.ai/models/mai-thinking-1)

Compare with

Phi-4-reasoning27.1

More from Microsoft

All models
Fara1.5-27BUnrankedFara1.5-4BUnrankedMAI-Cyber-1-FlashUnrankedMAI-Code-1.1-FlashUnrankedVibeVoice-ASR-Streaming 1.5BUnranked