← Gemini 3.1 Pro · API · High thinking
Evidenceacceptedthird_party_leaderboard
Chatbot Arena / LMArena human preference
Chatbot Arena mirror (openlm.ai) — Gemini-3.1-Pro
ev_2026_07_gem31pro_arena
Result
1505
Confidence B
- Tested
- 2026-07-11
- Runs
- 1
Configuration
How it was measured
- Exact model
- gemini-3.1-pro-preview
- Reasoning
- high
- Agent harness
- none
- Tool access
- chat product surface
- Prompting
- anonymous pairwise battles (Arena)
PRIMARY CHA — blind preference Elo (openlm Chatbot Arena mirror). Strong vision Elo supports multimancer product story (not a substitute for MMMU).
Caveats
Known limitations
- Preference Elo ≠ precise constraint following.
- Vision Elo is preference signal, not formal UI/MMMU accuracy.
- Mirror site; confirm LMArena at freeze.
Weighting
Evidence weight factors
The engine weights each evidence row by these factors (0–1) when fusing scores — higher is more trusted.
relevance0.90
quality0.85
recency1.00
health0.85
comparability0.75
reliability0.80
Benchmark
Chatbot Arena / LMArena human preference
Blind pairwise human preference battles aggregated into Elo ratings for chat quality, style, and helpfulness under real user prompts.
Benchmark source →