llmboard.ai
Benchmarks
CompareRankings
llmboard.ai
Benchmarks
CompareRankings
HomeModelsMAI Thinking 1

Microsoft model product

MAI Thinking 1

MAI-Thinking-1 is Microsoft AI's first in-house reasoning model, a 35B-active / ~1T-total parameter sparse Mixture of Experts model (base model MAI-Base-1) trained from scratch without distillation from third-party models. Built with Microsoft's Hill-Climbing Machine pipeline, it was pre-trained on 30T tokens of clean, commercially licensed, human-generated data (plus 3.55T mid-training tokens), then post-trained via reinforcement learning across STEM, agentic coding, and helpfulness/safety specialists consolidated into a single model. It delivers strong mathematical reasoning and software-engineering performance for its weight class, going toe-to-toe with Claude Opus 4.6 on SWE-Bench Pro and reaching 97.0% on AIME 2025. It supports a 256k token context window, function calling, and developer instructions, and is preferred over Claude Sonnet 4.6 in blind human side-by-side evaluations.

Updated Aug 10, 2026. Default version: MAI-Thinking-1

Compare
LLMBoard score63.4MAI-Thinking-1
Coverage80%10 benchmark families
Context windowN/ATokens
Official input priceN/AOfficial price unavailable

On this page

  • Specification
  • Capability
  • Benchmarks
  • Arena
  • Pricing
  • Versions
  • About
  • Compare
  • Similar models
  • FAQ

Model specification

Structured fields from the published default version.

Version
MAI-Thinking-1
Released
Jun 2, 2026
Knowledge cutoff
Unknown
Parameters
1T
Context window
N/A
Max output
N/A
Inputs
text
Outputs
text
Open weights
No
License
Proprietary

Capability profile

This profile uses the latest version under this unique model that has a calculated LLMBoard score. Arena and price are excluded.

MAI-Thinking-1 category scores

Benchmark results

Published benchmark records for the scored version MAI-Thinking-1.

23 rows
Columns

Show columns

AdvancedIF0.812100.0%CAug 7, 2026
AIR-Bench0.911100.0%CAug 7, 2026
CorpusQA0.811100.0%CAug 7, 2026
CyberSecEval 40.611100.0%CAug 7, 2026
GraphWalks0.913100.0%CAug 7, 2026
LongFact1.011100.0%CAug 7, 2026
SimpleQA Verified0.311100.0%CAug 7, 2026
TruthfulQA0.9118100.0%CAug 7, 2026
BFCL-v30.741983.3%CAug 7, 2026
AIME 20260.951775.0%CAug 7, 2026
LiveCodeBench v60.965390.4%CAug 7, 2026
LongBench v20.671762.5%CAug 7, 2026
HMMT Feb 260.881130.0%CAug 7, 2026
HealthBench Professional0.3990.0%CAug 7, 2026
MedXpertQA0.491227.3%CAug 7, 2026
AIME 20251.01511487.6%CAug 7, 2026
Multi-Challenge0.5162946.4%CAug 7, 2026
IFBench0.7192833.3%CAug 7, 2026
MMLU-Pro0.82112984.4%CAug 7, 2026
SWE-Bench Pro0.5374416.3%CAug 7, 2026
SWE-Bench Verified0.74110461.2%CAug 7, 2026
Terminal-Bench 2.00.5424914.6%CAug 7, 2026
GPQA0.86023374.6%CAug 7, 2026

Arena results

Preference and agent-evaluation signals from published Arena datasets.

No published Arena match

The default version has no published Arena rows, or its source alias has not been resolved.

Pricing

Official vendor API PAYG pricing is summarized first. The table then lists individual provider offerings without treating their minimum as the official price.

Official API
N/A
Official provider
N/A
Lowest third-party
N/A
Tracked offerings
0
No published price snapshot

The default version has no provider offering with current input or output token prices.

Official prices use only the vendor's configured official Provider and positive standard USD PAYG rates. Third-party offers remain explicitly labeled.

MAI Thinking 1 versions

All published versions linked to this unique model. The score columns identify the version used by the current overall ranking.

1 rows
Columns

Show columns

MAI-Thinking-1Jun 2, 202663.41TN/AN/ANoProprietary

What is MAI Thinking 1?

A concise description based on the published model registry.

MAI-Thinking-1 is Microsoft AI's first in-house reasoning model, a 35B-active / ~1T-total parameter sparse Mixture of Experts model (base model MAI-Base-1) trained from scratch without distillation from third-party models. Built with Microsoft's Hill-Climbing Machine pipeline, it was pre-trained on 30T tokens of clean, commercially licensed, human-generated data (plus 3.55T mid-training tokens), then post-trained via reinforcement learning across STEM, agentic coding, and helpfulness/safety specialists consolidated into a single model. It delivers strong mathematical reasoning and software-engineering performance for its weight class, going toe-to-toe with Claude Opus 4.6 on SWE-Bench Pro and reaching 97.0% on AIME 2025. It supports a 256k token context window, function calling, and developer instructions, and is preferred over Claude Sonnet 4.6 in blind human side-by-side evaluations.

Use the benchmark, Arena and pricing sections above as separate evidence. A missing field means the current data snapshot does not support that claim.

Data snapshot: 2026-08-07. Editorial model content is not available in the backend.

MAI Thinking 1 vs nearby models

Open a comparison with the three ranked models immediately above and below this model.

MAI Thinking 1vsClaude Sonnet 3.7MAI Thinking 1vsGemini 2.5 FlashMAI Thinking 1vsNova ProMAI Thinking 1vso1MAI Thinking 1vsLlama 3.3 Nemotron Super 49BMAI Thinking 1vsLlama 3.1 Nemotron Ultra 253B

Models similar to MAI Thinking 1

Recommendations prioritize the same model type and family, then the closest published LLMBoard score.

#83-13.0
MI

Phi 4 Reasoning Plus

Microsoft

50.3 LLMBoard

DetailsCompare
#92-15.5
MI

Phi 4 Reasoning

Microsoft

47.9 LLMBoard

DetailsCompare
#107-21.0
MI

Phi 3.5 MoE

Microsoft

42.4 LLMBoard

DetailsCompare
#121-26.4
MI

Phi 4

Microsoft

36.9 LLMBoard

DetailsCompare
#134-31.9
MI

Phi 4 Mini

Microsoft

31.5 LLMBoard

DetailsCompare
#144-37.5
MI

Phi 3.5 mini

Microsoft

25.9 LLMBoard

DetailsCompare

FAQ

Common questions about MAI Thinking 1.

When was MAI Thinking 1 released?

MAI Thinking 1's default version was released on Jun 2, 2026.

How much does MAI Thinking 1 cost?

No official standard PAYG price is currently available for MAI Thinking 1.

Who created MAI Thinking 1?

MAI Thinking 1 is published under Microsoft in the model registry.

What is the context window for MAI Thinking 1?

The current registry does not publish a context window for the default version.

Is MAI Thinking 1 open weight?

No. The default version is not marked as having publicly available weights.

How many API providers offer MAI Thinking 1?

No published provider offering is currently linked to the default version.

What models should I compare MAI Thinking 1 with?

Nearby ranked alternatives include Claude Sonnet 3.7, Gemini 2.5 Flash, Nova Pro.

Rankings

OverallCodingText ArenaPricing

Modalities

Image GenerationVideo GenerationSpeech-to-TextEmbeddings

Benchmarks

All BenchmarksReasoningMathCoding

Vendors

All VendorsOpenAIAnthropicGoogle
llmboard.aiCopyright 2026 llmboard.ai