Model evidence profile

MAI-Thinking-1

Microsoft · Proprietary · Reasoning

currentestimated
Overall public score
50.59
Source rank
#134
Evidence coverage
6 benchmarks · 1 sources

Strongest published evidencePublic overall score 50.59 at source rank #134.

Validate before choosingValidate current route pricing and evidence availability before choosing.

Revision benchmark_3b14dc55ef5307cfe91356d3ea348101 · Published Aug 30, 2026 · Checked Aug 30, 2026 · stale

Relative field position

Capability radar

Percentiles use eligible source ranks. Missing axes remain unavailable and never become zero.

Capability ranking percentile radarAgenticCodingInstruction FollowingKnowledgeMathMultimodal groundedOverallReasoning
Agentic percentile
56.1 percentileRank #62 of 140
Coding percentile
52.1 percentileRank #70 of 145
Instruction Following percentile
100.0 percentileRank #1 of 42
Knowledge percentile
77.8 percentileRank #13 of 55
Math percentile
Math: Unavailable
Multimodal grounded percentile
Multimodal grounded: Unavailable
Overall percentile
Overall: Unavailable
Reasoning percentile
Reasoning: Unavailable

Published measurements

Category scores

Scores retain their source units; percentile and rank appear only for eligible fields.

Agenticestimated
50.5
Rank
#62 of 140
Percentile
56.1%
Benchmarks
6
Codingestimated
51.5
Rank
#70 of 145
Percentile
52.1%
Benchmarks
6
Instruction Followingestimated
97.8
Rank
#1 of 42
Percentile
100.0%
Benchmarks
6
Knowledgeestimated
80.7
Rank
#13 of 55
Percentile
77.8%
Benchmarks
6
Mathestimated
73.5
Rank
Not ranked
Percentile
Unavailable
Benchmarks
6
Overallestimated
50.6
Rank
#134
Percentile
Unavailable
Benchmarks
6

Route-specific facts

Pricing and specifications

Conflicting routes remain separate and attributable.

Direct API pricing is unavailable.

Context window
256,000
Maximum output
Unavailable
Input modalities
Unavailable
Output modalities
Unavailable
Release date
2026-06-02
Self hosting
Not verified

Auditable evidence

Benchmark ledger

Display values, source ranks, and provenance remain visible without implying unsupported aggregate weight.

agentic

ScoreRankWeightLast UpdatedSource
50.47#62Not publishedAug 29, 2026BenchLM

coding

ScoreRankWeightLast UpdatedSource
51.54#70Not publishedAug 29, 2026BenchLM

instructionFollowing

ScoreRankWeightLast UpdatedSource
97.80#1Not publishedAug 29, 2026BenchLM

knowledge

ScoreRankWeightLast UpdatedSource
80.70#13Not publishedAug 29, 2026BenchLM

math

ScoreRankWeightLast UpdatedSource
73.50Not rankedNot publishedAug 29, 2026BenchLM

overall

ScoreRankWeightLast UpdatedSource
50.59#134Not publishedAug 29, 2026BenchLM

Related evidence

Compare this model