Artificial Analysis Launches VIALS-AA Leaderboard for Life Science Visual Reasoning
On November 7, Artificial Analysis launched the VIALS-AA leaderboard for Handshake AI's VIALS benchmark, evaluating frontier models' visual reasoning in life sciences and biotechnology. Gemini 4 tops the rankings, while gaps between models in cost and consistency are striking.
Confirmed
- The VIALS benchmark comprises 161 tasks across 8 life sciences and biotech domains, requiring models to interpret experimental images such as gel blots, flow cytometry plots, plasmid maps, and phylogenetic trees.
- The leaderboard shows Gemini 4 in first place; Claude's per-task cost is about 29x higher.
- Sol/Terra achieves visual reasoning performance close to frontier models at roughly 1/20 the cost, underscoring the price-performance gap.
- In consistency evaluations (running the same task multiple times), Kimi K3.1 Pro answered correctly in all three runs only 45% of the time, trailing GPT-6 Astra's 63%; Kimi K3.1 Pro got at least one correct answer on 119 of the 161 tasks.
- Similar overall scores can stem from different reasoning strategies: Claude Opus 5.5 (max with fallback) averages about 32 tool calls per task, while GPT-6 Astra averages only about 8.
Why it matters
- Artificial Analysis emphasizes that getting an answer right once does not mean getting it right reliably — consistency is treated as a key dimension of advantage for frontier models.
- The huge differences in cost and tool-call counts show that scores alone cannot capture a model's real-world cost and behavioral patterns; VIALS-AA offers a more well-rounded reference for model selection in the life sciences.
2026-11-07 ~ 2026-11-07 · 5 related posts
Primary sources
- VIALS-AA launches: Sol/Terra hit near-frontier scores at ~1/20th the cost per task of rivals — ArtificialAnlys ·
- New VIALS-AA leaderboard: Gemini 4 leads at 64%, Claude costs 29x more per task — ArtificialAnlys ·
- Consistency separates frontier models: Kimi K3.1 Pro passes 45% of VIALS tasks 3/3 vs Astra's 63% — ArtificialAnlys ·
- [source] New VIALS-AA leaderboard: Gemini 4 leads at 64%, Claude costs 29x more per task — ArtificialAnlys · 2026-11-07
- [source] VIALS-AA launches: Sol/Terra hit near-frontier scores at ~1/20th the cost per task of rivals — ArtificialAnlys · 2026-11-07
- Similar VIALS scores, opposite strategies: Opus 5.5 makes ~32 tool calls per task vs Astra's ~8 — ArtificialAnlys · 2026-11-07
- [source] Consistency separates frontier models: Kimi K3.1 Pro passes 45% of VIALS tasks 3/3 vs Astra's 63% — ArtificialAnlys · 2026-11-07
- Consistency test: Kimi K3.1 Pro passes 45% of VIALS tasks on all 3 tries vs 63% for GPT-6 Astra — ArtificialAnlys · 2026-11-07