VIALS-AA - Pass@3: leaderboard
Artificial Analysis' implementation of Handshake AI's VIALS (life-science image interpretation: blots, plasmid maps, phylogenetics, flow cytometry, molecular and protein structures). AA evaluates the released 161-task set; models get the question and image file paths, a sandbox with Python and no network, and an image-viewing tool if vision-capable, through AA's Stirrup harness. Three independent repeats per task, graded by GPT-6 Luna (medium) with the upstream judge prompts; a missing answer or refusal scores zero. This board reports share of tasks answered correctly at least once in three repeats. Distinct from the paper's direct-VLM 'VIALS' board. Rows keep AA's effort labels (for example Max, High, Xhigh).
Metric: Pass@3 (%). Source: artificialanalysis.ai. Saturation forecast: Around 2030. 19 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K3.1 Pro | 73.91 | #18 |
| 2 | Muse Wonder (Max) | 73.91 | #16 (Muse Wonder) |
| 3 | GPT-6 Astra (Max) | 73.29 | #5 (GPT-6 Astra) |
| 4 | Gemini 4 Argon (High) | 72.67 | #9 (Gemini 4 Argon) |
| 5 | Qwen 3.8 Max (0902) | 72.67 | #37 |
| 6 | Claude Haiku 5.5 (Max) | 71.43 | #46 (Claude Haiku 5.5) |
| 7 | GPT-6.1 Terra (Max) | 71.43 | #14 (GPT-6.1 Terra) |
| 8 | MiMo-V2.6-Pro | 70.81 | #51 |
| 9 | GPT-6.1 Sol (Max) | 70.81 | #6 (GPT-6.1 Sol) |
| 10 | MiniMax-M3.1-Flash | 70.81 | #57 |
| 11 | Gemini 4 Carbon (High) | 70.81 | #1 (Gemini 4 Carbon) |
| 12 | GLM-5.3 Flash | 67.08 | #79 |
| 13 | DeepSeek V4.1 Flash (1017) (Max) | 66.46 | #28 (DeepSeek V4.1 Flash (1017)) |
| 14 | Grok 4.7 (xHigh) | 65.84 | #26 (Grok 4.7) |
| 15 | Gemini 3.8 Flash (High) | 63.98 | #30 (Gemini 3.8 Flash) |
No result here: #3 Claude Opus 5.5, #10 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=vials-aa-pass-3 · How It Works · Data refreshed daily, snapshot 2026-11-07.