LMArena Agent Arena
LMArena · agent-2026-08-31-011508720696 · Outcome-based measurements from real LMArena agent sessions.
Model Ranking
Source | |||||||
|---|---|---|---|---|---|---|---|
| #1 | Claude Opus 5Anthropic | 11.61 | — | 17,073 | Claude Opus 5 (Max); 95% CI [9.58391463, 13.64552719]; sessions 17073; observations 2093237; rank 2unknown | Original sourcewinner eligible | Third-party benchmark |
| #2 | Claude Fable 5Anthropic | 10.58 | — | 34,905 | Claude Fable 5 (High); 95% CI [9.03695952, 12.11501267]; sessions 34905; observations 1709479; rank 3unknown | Original sourcewinner eligible | Third-party benchmark |
| #3 | GPT-5.6 SolOpenAI | 9.76 | — | 28,524 | GPT 5.6 Sol (xHigh); 95% CI [8.23209281, 11.28769899]; sessions 28524; observations 2917301; rank 4unknown | Original sourcewinner eligible | Third-party benchmark |
| #4 | Claude Sonnet 5Anthropic | 7.48 | — | 27,467 | Claude Sonnet 5 (High); 95% CI [5.31247081, 9.65098366]; sessions 27467; observations 2696236; rank 8unknown | Original sourcewinner eligible | Third-party benchmark |
| #5 | Qwen3.8-MaxQwen | 6.01 | — | 18,386 | Qwen3.8 Max; 95% CI [4.88117795, 7.13339178]; sessions 18386; observations 1372080; rank 14unknown | Original sourcewinner eligible | Third-party benchmark |
| #6 | Grok 4.6xAISelected model | 5.76 | — | 15,090 | Grok 4.6 (xHigh); 95% CI [4.47058991, 7.04827601]; sessions 15090; observations 1574671; rank 16unknown | Original sourcewinner eligible | Third-party benchmark |
| #7 | GLM-5.3Z.ai | 3.82 | — | 41,022 | GLM 5.3 (Max); 95% CI [3.01733683, 4.62585790]; sessions 41022; observations 2958163; rank 20unknown | Original sourcewinner eligible | Third-party benchmark |
| #8 | GPT-5.6 TerraOpenAI | 2.92 | — | 16,778 | GPT 5.6 Terra (xHigh); 95% CI [1.74076196, 4.09705014]; sessions 16778; observations 953717; rank 23unknown | Original sourcewinner eligible | Third-party benchmark |
| #9 | GPT-5.6 LunaOpenAI | 2.13 | — | 16,383 | GPT 5.6 Luna (xHigh); 95% CI [0.89962762, 3.35915348]; sessions 16383; observations 1269520; rank 26unknown | Original sourcewinner eligible | Third-party benchmark |
| #10 | Gemini 3.7 FlashGoogle DeepMind | 0.96 | — | 24,413 | Gemini 3.7 Flash (High); 95% CI [0.05817495, 1.85682345]; sessions 24413; observations 1756405; rank 30unknown | Original sourcewinner eligible | Third-party benchmark |
| #11 | Hy3Tencent | -1.64 | — | 23,436 | Hy3; 95% CI [-2.95971591, -0.31896988]; sessions 23436; observations 811149; rank 37unknown | Original sourcewinner eligible | Third-party benchmark |
| #12 | Qwen3.7-PlusQwen | -2.70 | — | 18,748 | Qwen3.7 Plus; 95% CI [-4.00378740, -1.40029439]; sessions 18748; observations 834350; rank 38unknown | Original sourcewinner eligible | Third-party benchmark |
| #13 | Gemini 3.5 FlashGoogle DeepMind | -2.85 | — | 94,447 | Gemini 3.5 Flash (High); 95% CI [-3.63579810, -2.05541341]; sessions 94447; observations 4118270; rank 39unknown | Original sourcewinner eligible | Third-party benchmark |
| #14 | Gemini 3.1 ProGoogle DeepMind | -3.13 | — | 83,248 | Gemini 3.1 Pro Preview; 95% CI [-4.00806078, -2.24630473]; sessions 83248; observations 2709356; rank 40unknown | Original sourcewinner eligible | Third-party benchmark |
| #15 | Gemini 3.6 FlashGoogle DeepMind | -3.52 | — | 16,850 | Gemini 3.6 Flash (High); 95% CI [-4.67028409, -2.36170848]; sessions 16850; observations 910940; rank 41unknown | Original sourcewinner eligible | Third-party benchmark |
| #16 | Mistral-Medium-3.5-128BMistral AI | -6.12 | — | 7,677 | Mistral Medium 3.5; 95% CI [-7.78867951, -4.45821930]; sessions 7677; observations 248889; rank 46unknown | Original sourcewinner eligible | Third-party benchmark |
| #17 | Gemini 3.5 Flash-LiteGoogle DeepMind | -10.81 | — | 22,348 | Gemini 3.5 Flash Lite; 95% CI [-12.19929495, -9.42408695]; sessions 22348; observations 634592; rank 51unknown | Original sourcewinner eligible | Third-party benchmark |
| #18 | Gemini 3 FlashGoogle DeepMind | -11.24 | — | 83,413 | Gemini 3 Flash; 95% CI [-12.27770994, -10.19849473]; sessions 83413; observations 1925564; rank 52unknown | Original sourcewinner eligible | Third-party benchmark |
18 models ranked by the best winner-eligible current-version score when available, otherwise the best publisher-artifact score; highest first. Observed Sep 2, 2026. The model you came from is highlighted.
Methodology and Coverage
Primary Evidence
Publisher Artifacts
Questions
LMArena Agent Arena FAQs
What does LMArena Agent Arena measure?+
Outcome-based measurements from real LMArena agent sessions. Model Markets classifies it as an agentic benchmark and preserves the publisher's agent-2026-08-31-011508720696 release as a distinct comparison cohort.
How are models ranked on LMArena Agent Arena?+
Models are ordered by outcome score in score, with higher scores ranked first. Each model appears once; a verified winner-eligible current-version result takes precedence over a publisher-artifact-only result, and missing scores are not estimated.
Which model currently leads LMArena Agent Arena?+
Claude Opus 5 leads the current verified table with 11.61 score on agent-2026-08-31-011508720696. This is a benchmark-specific result, not a universal model-quality claim.
How many models have a published LMArena Agent Arena score?+
18 catalog models are published from 56 source rows. 37 source identities remain quarantined rather than guessed.
Can LMArena Agent Arena scores be compared with other benchmarks?+
Raw scores should be compared only within the same benchmark version, metric, and protocol. Model Markets normalizes eligible scores only for aggregate rankings, and groups models by an identical benchmark set before ranking them. View aggregate rankings →
Why might a model be missing from LMArena Agent Arena?+
A model remains absent when the publisher has no current result, the source model identity is unresolved, the evaluation protocol is incompatible, or the evidence cannot be verified. Model Markets does not substitute a provider claim or infer a score from a related model.
How current is the LMArena Agent Arena leaderboard?+
The current Model Markets snapshot was observed Sep 2, 2026 from LMArena artifacts. The exact publisher source and each retained result artifact are linked on this page.