Artificial Analysis launches VIALS-AA leaderboard; GPT-6.1 Sol and Terra score 59% and 60%

AIArtificial Analysis reports that Kimi K3.1 Pro solves 119 of 161 VIALS tasks at least once across three attempts, versus 118 for GPT-6 Astra. Requiring all three attempts to succeed, Kimi passes 73 tasks (45%) and Astra passes 101 (63%).

AIArtificial Analysis reports that Claude Opus 5.5 (max with fallback) and GPT-6 Astra score 61% and 63% on VIALS-AA despite different tool use. Opus averages about 32 tool calls per task, including about 8 Image Measurement calls, while Astra averages about 8 calls with about 0.4 Image Measurement calls. Image Viewing makes up about 51% of Astra's tool calls and about 42% of Opus's.

AIArtificial Analysis reports that GPT-6.1 Sol scores 59% on VIALS-AA at about $0.056 per task, and GPT-6.1 Terra scores 60% at about $0.073. Compared with Sol, GPT-6 Astra reaches 63% at roughly 5 times the cost per task, Gemini 4 Carbon 64% at roughly 23 times, and Claude Opus 5.5 (max with fallback) 61% at roughly 29 times.

AIArtificial Analysis launches VIALS-AA, a leaderboard for Handshake AI's VIALS benchmark of 161 visual reasoning tasks in life sciences and biotech research. Gemini 4 Carbon (high) leads with 64% pass^3, followed by GPT-6 Astra (max) at 63% and Claude Opus 5.5 (max with fallback) at 61%. Claude Opus 5.5 averages $1.64 per task, about 29 times the cost of GPT-6.1 Sol (max) at roughly $0.056.