Skip to main content

Vals MedQA (MedQA)

Evaluating language model bias in medical questions.

Data verified

How BenchLM shows MedQA

BenchLM mirrors the public Vals AI MedQA leaderboard captured from https://www.vals.ai/benchmarks/medqa and updated by Vals on April 16, 2026. The snapshot preserves overall scores, uncertainty, latency, cost-per-test metadata, and task-level scores where Vals publishes them.

MedQA is display only on BenchLM. Vals proprietary or Vals-hosted aggregate views are useful context, but BenchLM does not use them as weighted ranking inputs or as a replacement for benchmark-native source records.

95 Vals rows7 task viewspublic datasetTasks: Overall, Unbiased, Hispanic, Black, AsianDisplay only

MedQA score on MedQA — April 16, 2026

BenchLM mirrors the published medqa score view for MedQA. O1 leads the public snapshot at 96.52% , followed by GPT-5.1 (96.38%) and Gemini 3.1 Pro Preview (96.37%). BenchLM does not use these results to rank models overall.

95 modelsExternal benchmark mirrorsCurrentDisplay onlyUpdated April 16, 2026

The published MedQA snapshot is tightly clustered at the top: O1 sits at 96.52%, while the third row is only 0.15 points behind. The broader top-10 spread is 0.64 points, so many of the published scores sit in a relatively narrow band.

95 models have been evaluated on MedQA. The benchmark falls in the External benchmark mirrors category. BenchLM tracks this category separately from its weighted global scoring system, so these results are best compared on the dedicated Korean benchmark views. MedQA is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About MedQA

Year

2026

Tasks

Medical question answering

Format

Accuracy score

Difficulty

Medical knowledge and bias evaluation

BenchLM mirrors the public Vals AI MedQA leaderboard as display-only external evidence. The captured snapshot preserves overall scores, task-level scores where Vals publishes them, uncertainty, latency, and cost-per-test metadata. It is excluded from BenchLM weighted rankings.

BenchLM freshness & provenance

Version

MedQA 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

MedQA score table (95 models)

1
O1openai/o1-2024-12-17
96.52%
2
GPT-5.1openai/gpt-5.1-2025-11-13
96.38%
3
Gemini 3.1 Pro Previewgoogle/gemini-3.1-pro-preview
96.37%
4
GPT-5openai/gpt-5-2025-08-07
96.32%
5
GPT-5.4openai/gpt-5.4-2026-03-05
96.09%
6
O3openai/o3-2025-04-16
96.06%
7
GPT-5 Miniopenai/gpt-5-mini-2025-08-07
96.06%
8
Gemini 3 Pro Previewgoogle/gemini-3-pro-preview
96.03%
9
O4 Miniopenai/o4-mini-2025-04-16
96.02%
10
Claude Opus 4.5 20251101 Thinkinganthropic/claude-opus-4-5-20251101-thinking
95.88%
11
Gemini 3 Flash Previewgoogle/gemini-3-flash-preview
95.81%
12
Claude Opus 4.6 Thinkinganthropic/claude-opus-4-6-thinking
95.41%
13
Qwen3.5 Plus Thinkingalibaba/qwen3.5-plus-thinking
95.21%
14
O3 Miniopenai/o3-mini-2025-01-31
94.83%
15
Claude Sonnet 4.5 20250929 Thinkinganthropic/claude-sonnet-4-5-20250929-thinking
94.71%
16
Grok 4.20 0309 Reasoninggrok/grok-4.20-0309-reasoning
94.55%
17
Kimi K2.5 Thinkingkimi/kimi-k2.5-thinking
94.37%
18
GLM 5 Thinkingzai/glm-5-thinking
94.27%
19
GPT-5.2openai/gpt-5.2-2025-12-11
94.13%
20
DeepSeek V3p2 Thinkingfireworks/deepseek-v3p2-thinking
93.92%
21
GLM 4.7zai/glm-4.7
93.74%
22
Claude Opus 4.1 20250805 Thinkinganthropic/claude-opus-4-1-20250805-thinking
93.59%
23
GPT-5 Nanoopenai/gpt-5-nano-2025-08-07
93.26%
24
Claude Opus 4.5anthropic/claude-opus-4-5-20251101
93.16%
25
Gemini 2.5 Pro Exp 03 25google/gemini-2.5-pro-exp-03-25
93.14%
26
O1 Previewopenai/o1-preview-2024-09-12
93.01%
27
Claude Opus 4anthropic/claude-opus-4-20250514
92.87%
28
Claude Sonnet 4 20250514 Thinkinganthropic/claude-sonnet-4-20250514-thinking
92.71%
29
Kimi K2 Thinkingkimi/kimi-k2-thinking
92.59%
30
Claude Opus 4.1anthropic/claude-opus-4-1-20250805
92.53%
31
MiniMax M2.5minimax/MiniMax-M2.5
92.53%
32
Grok 4 0709grok/grok-4-0709
92.49%
33
Grok 2 1212grok/grok-2-1212
92.32%
34
GLM 4.6zai/glm-4.6
92.22%
35
Grok 4.1 Fast Reasoninggrok/grok-4-1-fast-reasoning
92.08%
36
Grok 4 Fast Reasoninggrok/grok-4-fast-reasoning
92.07%
37
Claude Sonnet 4.6anthropic/claude-sonnet-4-6
92.06%
38
Gemini 2.5 Flash Preview 09 2025google/gemini-2.5-flash-preview-09-2025
91.43%
39
GPT Oss 120bfireworks/gpt-oss-120b
91.36%
40
GPT-4.1openai/gpt-4.1-2025-04-14
91.18%
41
Gemini 2.5 Flash Preview 09 2025 Thinkinggoogle/gemini-2.5-flash-preview-09-2025-thinking
91.17%
42
MiniMax M2.1minimax/MiniMax-M2.1
91.16%
43
Gemini 2.5 Flash Preview 04 17 Thinkinggoogle/gemini-2.5-flash-preview-04-17-thinking
91.02%
44
DeepSeek R1fireworks/deepseek-r1
90.80%
45
Qwen3 235b A22bfireworks/qwen3-235b-a22b
90.62%
46
Claude Sonnet 4anthropic/claude-sonnet-4-20250514
90.35%
47
O1 Miniopenai/o1-mini-2024-09-12
90.22%
48
Claude 3.7 Sonnet 20250219 Thinkinganthropic/claude-3-7-sonnet-20250219-thinking
90.22%
49
Grok 3 Mini Fast High Reasoninggrok/grok-3-mini-fast-high-reasoning
90.10%
50
GLM 4.5zai/glm-4.5
89.97%
51
Magistral Medium 2509mistralai/magistral-medium-2509
89.47%
52
DeepSeek V3p2fireworks/deepseek-v3p2
89.45%
53
Gemini 2.5 Flash Lite Preview 09 2025 Thinkinggoogle/gemini-2.5-flash-lite-preview-09-2025-thinking
88.87%
54
Grok 3 Mini Fast Low Reasoninggrok/grok-3-mini-fast-low-reasoning
88.65%
55
Meta Llama Meta Llama 3.1 405B Instruct Turbotogether/meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo
88.24%
56
GPT-4oopenai/gpt-4o-2024-08-06
88.16%
57
Qwen3 Max Previewalibaba/qwen3-max-preview
87.38%
58
Qwen3 Maxalibaba/qwen3-max
87.37%
59
Gemini 2.5 Flash Preview 04 17google/gemini-2.5-flash-preview-04-17
86.73%
60
Meta Llama Meta Llama 3.1 70B Instruct Turbotogether/meta-llama/Meta-Llama-3.1-70B-Instruct-Turbo
84.78%
61
GPT-4.1 Miniopenai/gpt-4.1-mini-2025-04-14
84.63%
62
Moonshotai Kimi K2 Instructtogether/moonshotai/Kimi-K2-Instruct
83.97%
63
Grok 3grok/grok-3
83.85%
64
Claude 3.5 Sonnetanthropic/claude-3-5-sonnet-20241022
83.19%
65
GPT Oss 20bfireworks/gpt-oss-20b
82.88%
66
Magistral Small 2509mistralai/magistral-small-2509
82.36%
67
Mistral Large 2512mistralai/mistral-large-2512
82.23%
68
DeepSeek V3 0324fireworks/deepseek-v3-0324
82.00%
69
GPT-4 Turboopenai/gpt-4-turbo
81.99%
70
Gemini 2.0 Flash 001google/gemini-2.0-flash-001
81.47%
71
DeepSeek V3fireworks/deepseek-v3
80.90%
72
Command A 03 2025cohere/command-a-03-2025
80.55%
73
Gemini 2.5 Flash Lite Preview 09 2025google/gemini-2.5-flash-lite-preview-09-2025
80.33%
74
Claude Haiku 4.5 20251001 Thinkinganthropic/claude-haiku-4-5-20251001-thinking
79.57%
75
Mistral Medium 2505mistralai/mistral-medium-2505
78.23%
76
Qwen Qwen2.5 72B Instruct Turbotogether/Qwen/Qwen2.5-72B-Instruct-Turbo
77.39%
77
Gemini 1.5 Pro 002google/gemini-1.5-pro-002
76.53%
78
Mistral Large 2411mistralai/mistral-large-2411
76.22%
79
Grok 4.1 Fast Non Reasoninggrok/grok-4-1-fast-non-reasoning
76.03%
80
Grok 4 Fast Non Reasoninggrok/grok-4-fast-non-reasoning
75.36%
81
GPT-4o Miniopenai/gpt-4o-mini-2024-07-18
72.44%
82
Mistral Small 2503mistralai/mistral-small-2503
69.10%
83
GPT-4.1 Nanoopenai/gpt-4.1-nano-2025-04-14
68.22%
84
Jamba 1.5 Largeai21labs/jamba-1.5-large
68.11%
85
Meta Llama Meta Llama 3.1 8B Instruct Turbotogether/meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo
62.61%
86
Mistralai Mixtral 8x22B Instruct V0.1together/mistralai/Mixtral-8x22B-Instruct-v0.1
62.14%
87
GPT-3.5 Turboopenai/gpt-3.5-turbo
58.47%
88
Mistral Small 2402mistralai/mistral-small-2402
56.98%
89
Jamba 1.5 Miniai21labs/jamba-1.5-mini
55.18%
90
Mistralai Mixtral 8x7B V0.1together/mistralai/Mixtral-8x7B-v0.1
53.22%
91
Jamba Mini 1.6ai21labs/jamba-mini-1.6
52.52%
92
Meta Llama Llama 4 Scout 17B 16E Instructtogether/meta-llama/Llama-4-Scout-17B-16E-Instruct
50.90%
93
Jamba Large 1.6ai21labs/jamba-large-1.6
50.70%
94
Llama4 Maverick Instruct Basicfireworks/llama4-maverick-instruct-basic
43.30%
95
Command R Pluscohere/command-r-plus
2.65%

FAQ

What does MedQA measure?

Evaluating language model bias in medical questions.

Which model leads the published MedQA snapshot?

O1 currently leads the published MedQA snapshot with 96.52% medqa score. BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many models are evaluated on MedQA?

95 AI models are included in BenchLM's mirrored MedQA snapshot, based on the public leaderboard captured on April 16, 2026.

Last updated: April 16, 2026 · mirrored from the public benchmark leaderboard

The AI models change fast. We track them for you.

For engineers, researchers, and the plain curious — a weekly brief on new models, ranking shifts, and pricing changes.

Free. No spam. Unsubscribe anytime.