AIEvals

Benchmarks

MLCR-AA

Synthesis of fragmented medical records spanning roughly 25k–64k tokens.

Medical reasoningWisedocs / Artificial Analysis70 models run by Artificial AnalysisUpdated 30 September 2026

How it is run

AA’s private hard subset, three repeats. Three judges must find a concise response accurate and complete by majority vote; overlong or empty answers fail.

What it cannot tell you

Synthetic records and a private hard subset. Overall pass rate differs from accuracy among judged responses.

Leader
Claude Sonnet 5.575.00% · Artificial Analysis results, grouped here by release
Top-ten spread
31.67%Leader minus tenth place, or last place in smaller fields.
Median
15.84%of the 70 rows published
No published result
9of 46 models on this site

Artificial Analysis results by release

Highest published effort per model release, grouped here from AA’s public results. Grouping ranks are derived here; source settings and model-publisher reports stay separate.

#ModelScoreEffort
1Claude Sonnet 5.575.00%max
2Claude Fable 5.171.11%max
3Claude Opus 5.566.67%max
4Claude Fable 564.44%max
5Claude Opus 555.56%max
6Claude Sonnet 555.00%max
7GLM 5.3 Flash51.11%max
8GLM-5.348.33%max
9Claude Opus 4.845.56%max
10Muse Spark 1.343.33%max
11Kimi K338.33%max
12GPT-6 Astra35.00%max
13GPT-6.1 Sol33.89%max
14GPT-5.6 Terra31.67%max
15Muse Spark 1.231.11%xhigh
16Claude Opus 4.728.89%high
17GPT-5.6 Sol26.11%max
18Claude Sonnet 4.624.44%max
19DeepSeek V4.1 Flash22.78%max
20Gemini 3.8 Flash21.67%high
20Qwen3.8 27B21.67%xhigh
22DeepSeek V4 Pro 042421.11%max
23Muse Glimmer20.00%high
23Qwen3.8 Max (0902)20.00%Not specified
25GPT-5.6 Luna19.44%max
25Qwen3.8 Max19.44%Not specified
27Gemini 3.5 Flash18.33%high
27MiMo-V2.6-Pro18.33%Not specified
29GPT-5.517.78%xhigh
29DeepSeek V4 Pro 081317.78%max
31MiniMax-M317.22%Not specified
32Step 5 Preview16.67%Not specified
33GPT-6 Luna16.11%max
33Kimi K2.7 Code16.11%Not specified
33GPT-6 Sol16.11%max
36Gemini 3.1 Pro Preview15.56%Not specified
37Grok 4.715.00%xhigh
37Grok 4.515.00%high
37DeepSeek V4 Flash 042015.00%max
37Gemini 3.7 Flash15.00%high
41Gemini 3.6 Flash14.44%high
42DeepSeek V4 Flash 073113.33%max
43Inkling12.22%xhigh
43Grok 4.612.22%high
45Gemini 3 Flash Preview11.67%Not specified
46Nemotron 3 Ultra 550B A55B11.11%Not specified
47Grok 4.310.00%high
48MiMo-V2.5-Pro9.44%Not specified
48Qwen3.7 Plus9.44%Not specified
50Gemini 2.5 Flash8.89%Not specified
51Gemini 3.5 Flash-Lite7.22%high
51GLM-5.27.22%max
53Gemini 3.1 Flash-Lite6.11%high
53Claude 4.5 Haiku6.11%Not specified
55GPT-5.4 nano5.00%xhigh
56Nova 2.0 Pro Preview3.89%medium
56GPT-5.4 mini3.89%xhigh
58MiniMax-M2.73.33%Not specified
58Nemotron 3 Super 120B A12B3.33%Not specified
60Llama 4 Maverick2.22%Not specified
60Gemini 2.5 Flash-Lite2.22%Not specified
62Mistral Medium 3.51.67%high
63Nemotron 3.5 Lightning1.11%Not specified
63gpt-oss-120b1.11%high
65Command A+0.56%Not specified
65gpt-oss-20b0.56%high
65Solar Pro 30.56%high
68Nova Lite0.00%Not specified
68Qwen3.8 2.4T A95B0.00%Not specified
68GPT-4o mini0.00%Not specified
Source: artificialanalysis.ai/evaluations/mlcr-aa

No published result tracked

9 of the 46 models on this site have no published number on this benchmark, so their cells in the home table stay blank rather than reading as zero.

Qwen3.8-Flash-NextMiMo-V2.6-FlashGPT-5.4DeepSeek V4.1 Flash 1017GLM-LegalClaude Haiku 5.5BeamGemini 4 ArgonHunyuan Hy4 Preview

Sources

Every page used for this benchmark, read on 17 October 2026.