XNews

Artificial Analysis Launches VIALS-AA Leaderboard for Life Sciences Visual Reasoning

Artificial Analysis Launches VIALS-AA Leaderboard for Life Sciences Visual Reasoning

Artificial Analysis launched VIALS-AA, a leaderboard for Handshake AI's VIALS benchmark, which tests visual reasoning in life sciences and biotech across 161 tasks in eight domains. Gemini 4 Carbon (high) leads at 64%, followed by GPT-6 Astra (max) at 63% and Claude Opus 5.5 (max with fallback) at 61%, while Kimi K3.1 Pro tops open weights models at 45%.

Event version 1.

Key insights

Top three models separated by 3 points

Gemini 4 Carbon (high) from Google DeepMind leads with 64%, followed by GPT-6 Astra (max) from OpenAI at 63% and Claude Opus 5.5 (max with fallback) from Anthropic at 61%.

Cost per task varies widely

Claude Opus 5.5 (max with fallback) averages $1.64 per task, about 29 times as much as GPT-6.1 Sol (max) at roughly $0.056. GPT-6 Astra (max) and GPT-6.1 Sol (max) average about 8 turns per task, compared with about 27 for Claude Opus 5.5 (max with fallback) and about 25 for Claude Sonnet 5.5 (max with fallback).

Open weights models trail by about 19 points

Kimi K3.1 Pro leads open weights models on VIALS-AA at 45%, narrowly ahead of MiMo-V2.6-Pro at 43%.

Safety refusals on 14 tasks

14 of 161 tasks saw safety refusals on at least one model. Claude Sonnet 5.5 (max with fallback) refused on 6 tasks, the most of any model, and half of the 14 affected tasks saw refusals from multiple models.

Evaluation method

Each task is run three times in Stirrup, an open source agent harness with code execution and image inspection tools. GPT-6 Luna (medium) grades each answer against the reference answer, and the share of tasks judged correct in all three repeats (pass^3) is reported.

Source posts