Clinical Benchmarks
Menu

MAI-Thinking-1

Released 12 Aug 2026$2 input, $8 output per million tokens256K context1.0TproprietaryAlso written as MAI Thinking

1 result on 1 of 13 boards, latest measured Aug 2026. Compare with other models

Clinical Benchmarks Index14.3Rank 141 of 1491 of 10 boardsAverage × multiplier 35.0 × 0.408 = 14.3This site’s own calculation. Working How it is computed

Results

Each line sits on its own board's scale. Tall tick: the board leader.

Clinical reasoning and knowledge

  1. HealthBench Professional
    length-adjusted (HealthBench Professional length penalty), standard GPT-5.4 grader and OpenAI rubrics, Microsoft AI run; printed at integer precision.
    0.350
    Rank 34 of 35Leader GPT-6 Astra (Anthropic run) 0.703
    Vendor-reported
    Measured Aug 2026

Index working

The Clinical Benchmarks Index is this site's own calculation: a weighted average of the board scores times a multiplier for the boards behind it. How it is computed

  • HealthBench Professional0.35035.0×0.743length-adjusted (HealthBench Professional length penalty), standard GPT-5.4 grader and OpenAI rubrics, Microsoft AI run; printed at integer precision.
One board, so its score
35.0
Multiplier for 1 of 10 boards
× 0.408
Index
35.0 × 0.408 = 14.3

Score: the result on the board’s own 0 to 100 scale. Dot: this model; tick: the board’s leader. Weight: how much the board counts in the weighted average. Each result links to its quote and page; the line under it is the setting it was measured with.

Sources

The quote, page and document for each of the 1 result
  1. 34HealthBench Professional length-adjusted 0.350
    Printed as 35Vendor-reported, measured Aug 2026Configuration: length-adjusted (HealthBench Professional length penalty), standard GPT-5.4 grader and OpenAI rubrics, Microsoft AI run; printed at integer precision.
    MAI-Thinking-1: Building a Hill-Climbing Machine model card, Microsoft AI, 12 Aug 2026. p. 54, Table 12 'Post-trained model evaluation results on various public benchmarks', Health group, column HealthBench Prof.; protocol Appendix K.6 p. 106: HealthBench Professional introduces a length penalty for the primary metric, to correct for a well-observed correlation between lengthy responses and artificially increased LLM-grader scores. For all reported scores, we use the standard GPT-5.4 grader and rubrics provided by OpenAI.
    Model AIR-Bench CyberSec Instruct CyberSec Auto Long Fact Truthful QA HealthBench Prof. MedXpert QA MAI-Thinking-1 88 63 63 98 88 35 43 Sonnet 4.6 88 62 56 98 88 38 49
    Every result from this document