AutoMedBench Lite
Leaderboard
Average model performance across all seven released Lite tracks.
Leaderboard
7-track average
Overall = 1/2 Agentic + 1/2 Task
| Rank | Model | Parameter size | Release | Overall | Agentic | Task | Avg cost | Avg time | Avg turns |
|---|---|---|---|---|---|---|---|---|---|
| 1 |
Claude Opus 4.6Anthropic
|
~1T | 2026-02-05 | 77.23 | 89.10 | 65.36 | $6.57 | 18.2 min | 18.3 |
| 2 |
Gemini 3.1 ProGoogle
|
~1T | 2026-02-19 | 75.21 | 83.49 | 66.93 | $3.85 | 18.0 min | 18.1 |
| 3 |
GLM-5Z.ai
|
744B | 2026-02-12 | 72.86 | 89.35 | 56.38 | $1.29 | 17.8 min | 31.3 |
| 4 |
Qwen3.5-397BAlibaba Cloud
|
397B | 2026-02-15 | 69.87 | 80.46 | 59.27 | $1.01 | 20.8 min | 25.1 |
| 5 |
MiniMax M2.5MiniMax
|
230B | 2026-02-12 | 66.31 | 77.71 | 54.91 | $1.36 | 20.6 min | 28.5 |
| 6 |
ChatGPT 5.4OpenAI
|
~1T | 2026-03-05 | 66.09 | 76.76 | 55.43 | $2.36 | 16.7 min | 10.4 |
| 7 |
Qwen3.5-9BAlibaba Cloud
|
9B | 2026-02-27 | 19.87 | 11.33 | 28.41 | free |
19.3 min | 52.4 |
| 8 |
Qwen3.5-4BAlibaba Cloud
|
4B | 2026-02-27 | 6.05 | 12.10 | 0.00 | free |
13.6 min | 72.6 |
Scores and operational metrics are equal-weight averages across the seven Lite release tasks. Cost is USD; time is wall-clock; turns are conversational turns. ~ marks an estimate.