Benchmark Almanac

Agent Arena

Claude Opus 5.5 0.21 leads since Sep 2026 Only in system cards

Arena ranks models as the orchestrator in real Agent Mode sessions on arena.ai. Scores use behavioral signals such as file downloads, disapproval events, retries and steerability.

Arena says the board measures signals from real sessions rather than preference votes alone. It went live on 2026-06-04, and signal definitions changed on 2026-10-01 and 2026-10-12.

Made by Arena (formerly LMArena) Released Jun 2026 General assistants All tasks public Last in a system card Jul 2026 Last model added Oct 2026 Scores refreshed 2 Nov 2026
Level
Everyday Most adults, without training
Scored
Ranked by votes People pick the better of two answers, as in an arena, and the votes are turned into ratings.
Answers
Agent task Many steps in a terminal, browser, desktop, app, game or robot.
Paper Leaderboard arena.ai
54
Entries on its own leaderboard, 53 models we track
1
System card reports it, from 1 lab

Claude Opus 5.5 leads with 0.21

Best score per entry from the benchmark's own leaderboard, read on 2 Nov 2026. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.

54 entries. Scroll sideways for the rest.

  1. 0.21 Claude Opus 5.5
  2. 0.16 Gemini 4 Carbon
  3. 0.14 Claude Sonnet 5.5
  4. 0.13 Claude Fable 5.1
  5. 0.13 GPT-6 Astra
  6. 0.12 GPT-6.1 Sol
  7. 0.11 Gemini 4 Argon
  8. 0.11 GPT-6.1 Terra
  9. 0.10 Muse Wonder Full
  10. 0.08 Claude Fable 5
  11. 0.08 GPT-6 Sol
  12. 0.08 Claude Opus 5
  13. 0.06 Kimi K3.1 Pro
  14. 0.05 GPT-5.6 Sol
  15. 0.03 Muse Wonder
  16. 0.03 Claude Sonnet 5
  17. 0.03 MiMo-V2.6 Pro
  18. 0.03 DeepSeek V4.1 Flash
  19. 0.03 Gemini 3.8 Flash
  20. 0.03 Muse Spark 1.3
  21. 0.02 Grok 4.7
  22. 0.02 GPT-6 Luna
  23. 0.01 Claude Haiku 5.5
  24. 0.01 GLM-5.2
  25. 0.01 Tencent Hy4 preview
  26. 0.01 Kimi K3
  27. 0.01 MiMo-V2.6 Flash
  28. 0.01 GLM-5.3
  29. 0.00 Grok 4.6
  30. 0.00 Qwen 3.8 Max
  31. -0.00 DeepSeek-V4-Pro
  32. -0.01 Gemini 3.7 Flash
  33. -0.01 GPT-5.6 Terra
  34. -0.02 GLM-5.3-Flash
  35. -0.03 Step 5 Preview
  36. -0.04 MiniMax-M3.1-Flash
  37. -0.04 Qwen3.8-Flash-Next
  38. -0.04 Gemini 3.1 Pro
  39. -0.04 Qwen 3.8 27B
  40. -0.05 Beam 1.0
  41. -0.06 Mistral Large 4
  42. -0.07 Qwen3.7-Max
  43. -0.07 Tencent Hy3
  44. -0.07 Qwen3.7-Plus
  45. -0.08 Muse Spark 1.2
  46. -0.08 Muse Spark 1.1
  47. -0.09 MiMo-V2.5-Pro
  48. -0.10 MiniMax-M3
  49. -0.13 Mistral Medium 3.5
  50. -0.15 Nemotron 3.5 Super VL
  51. -0.16 Nemotron 3 Ultra
  52. -0.16 Inkling
  53. -0.17 Inkling-Small
  54. -0.21 Solar Pro 4
  1. 1 Claude Opus 5.5 max 0.21
  2. 2 Gemini 4 Carbon high 0.16
  3. 3 Claude Sonnet 5.5 max 0.14
  4. 4 Claude Fable 5.1 max 0.13
  5. 5 GPT-6 Astra max 0.13
  6. 6 GPT-6.1 Sol max 0.12
  7. 7 Gemini 4 Argon high 0.11
  8. 8 GPT-6.1 Terra max 0.11
  9. 9 Muse Wonder Full max 0.10
  10. 10 Claude Fable 5 high 0.08
  11. 11 GPT-6 Sol max 0.08
  12. 12 Claude Opus 5 max 0.08
  13. 13 Kimi K3.1 Pro max 0.06
  14. 14 GPT-5.6 Sol xhigh 0.05
  15. 15 Muse Wonder max 0.03
  16. 16 Claude Sonnet 5 high 0.03
  17. 17 MiMo-V2.6 Pro 0.03
  18. 18 DeepSeek V4.1 Flash max 0.03
  19. 19 Gemini 3.8 Flash high 0.03
  20. 20 Muse Spark 1.3 max 0.03
  21. 21 Grok 4.7 xhigh 0.02
  22. 22 GPT-6 Luna max 0.02
  23. 23 Claude Haiku 5.5 high 0.01
  24. 24 GLM-5.2 max 0.01
  25. 25 Tencent Hy4 preview 0.01
  26. 26 Kimi K3 max 0.01
  27. 27 MiMo-V2.6 Flash 0.01
  28. 28 GLM-5.3 max 0.01
  29. 29 Grok 4.6 xhigh 0.00
  30. 30 Qwen 3.8 Max 0.00
  31. 31 DeepSeek-V4-Pro high -0.00
  32. 32 Gemini 3.7 Flash high -0.01
  33. 33 GPT-5.6 Terra xhigh -0.01
  34. 34 GLM-5.3-Flash -0.02
  35. 35 Step 5 Preview -0.03
  36. 36 MiniMax-M3.1-Flash -0.04
  37. 37 Qwen3.8-Flash-Next -0.04
  38. 38 Gemini 3.1 Pro -0.04
  39. 39 Qwen 3.8 27B -0.04
  40. 40 Beam 1.0 -0.05
  41. 41 Mistral Large 4 -0.06
  42. 42 Qwen3.7-Max -0.07
  43. 43 Tencent Hy3 -0.07
  44. 44 Qwen3.7-Plus -0.07
  45. 45 Muse Spark 1.2 xhigh -0.08
  46. 46 Muse Spark 1.1 -0.08
  47. 47 MiMo-V2.5-Pro -0.09
  48. 48 MiniMax-M3 -0.10
  49. 49 Mistral Medium 3.5 -0.13
  50. 50 Nemotron 3.5 Super VL nvfp4 -0.15
  51. 51 Nemotron 3 Ultra -0.16
  52. 52 Inkling -0.16
  53. 53 Inkling-Small -0.17
  54. 54 Solar Pro 4 -0.21
#EntryScoreSettingDateSource
1 Claude Opus 5.5 0.21 max Sep 2026 Official leaderboard
2 Gemini 4 Carbon 0.16 high Oct 2026 Official leaderboard
3 Claude Sonnet 5.5 0.14 max Sep 2026 Official leaderboard
4 Claude Fable 5.1 0.13 max Sep 2026 Official leaderboard
5 GPT-6 Astra 0.13 max Sep 2026 Official leaderboard
6 GPT-6.1 Sol 0.12 max Sep 2026 Official leaderboard
7 Gemini 4 Argon 0.11 high Sep 2026 Official leaderboard
8 GPT-6.1 Terra 0.11 max Oct 2026 Official leaderboard
9 Muse Wonder Full 0.10 max Oct 2026 Official leaderboard
10 Claude Fable 5 0.08 high Jun 2026 Official leaderboard
11 GPT-6 Sol 0.08 max Sep 2026 Official leaderboard
12 Claude Opus 5 0.08 max Jul 2026 Official leaderboard
13 Kimi K3.1 Pro 0.06 max Oct 2026 Official leaderboard
14 GPT-5.6 Sol 0.05 xhigh Jul 2026 Official leaderboard
15 Muse Wonder 0.03 max Oct 2026 Official leaderboard
16 Claude Sonnet 5 0.03 high Jun 2026 Official leaderboard
17 MiMo-V2.6 Pro 0.03 Sep 2026 Official leaderboard
18 DeepSeek V4.1 Flash 0.03 max Sep 2026 Official leaderboard
19 Gemini 3.8 Flash 0.03 high Sep 2026 Official leaderboard
20 Muse Spark 1.3 0.03 max Sep 2026 Official leaderboard
21 Grok 4.7 0.02 xhigh Sep 2026 Official leaderboard
22 GPT-6 Luna 0.02 max Sep 2026 Official leaderboard
23 Claude Haiku 5.5 0.01 high Oct 2026 Official leaderboard
24 GLM-5.2 0.01 max Jun 2026 Official leaderboard
25 Tencent Hy4 preview 0.01 Aug 2026 Official leaderboard
26 Kimi K3 0.01 max Jul 2026 Official leaderboard
27 MiMo-V2.6 Flash 0.01 Sep 2026 Official leaderboard
28 GLM-5.3 0.01 max Aug 2026 Official leaderboard
29 Grok 4.6 0.00 xhigh Aug 2026 Official leaderboard
30 Qwen 3.8 Max 0.00 Aug 2026 Official leaderboard
31 DeepSeek-V4-Pro -0.00 high Apr 2026 Official leaderboard
32 Gemini 3.7 Flash -0.01 high Aug 2026 Official leaderboard
33 GPT-5.6 Terra -0.01 xhigh Jul 2026 Official leaderboard
34 GLM-5.3-Flash -0.02 Aug 2026 Official leaderboard
35 Step 5 Preview -0.03 Sep 2026 Official leaderboard
36 MiniMax-M3.1-Flash -0.04 Oct 2026 Official leaderboard
37 Qwen3.8-Flash-Next -0.04 Oct 2026 Official leaderboard
38 Gemini 3.1 Pro -0.04 Feb 2026 Official leaderboard
39 Qwen 3.8 27B -0.04 Aug 2026 Official leaderboard
40 Beam 1.0 -0.05 Oct 2026 Official leaderboard
41 Mistral Large 4 -0.06 Oct 2026 Official leaderboard
42 Qwen3.7-Max -0.07 May 2026 Official leaderboard
43 Tencent Hy3 -0.07 Jul 2026 Official leaderboard
44 Qwen3.7-Plus -0.07 Jun 2026 Official leaderboard
45 Muse Spark 1.2 -0.08 xhigh Aug 2026 Official leaderboard
46 Muse Spark 1.1 -0.08 Jul 2026 Official leaderboard
47 MiMo-V2.5-Pro -0.09 Apr 2026 Official leaderboard
48 MiniMax-M3 -0.10 Jun 2026 Official leaderboard
49 Mistral Medium 3.5 -0.13 Apr 2026 Official leaderboard
50 Nemotron 3.5 Super VL -0.15 nvfp4 Oct 2026 Official leaderboard
51 Nemotron 3 Ultra -0.16 Jun 2026 Official leaderboard
52 Inkling -0.16 Jul 2026 Official leaderboard
53 Inkling-Small -0.17 Jul 2026 Official leaderboard
54 Solar Pro 4 -0.21 Aug 2026 Official leaderboard

Labs report scores for 1 model in their cards. We plot every source in the charts below.

4 models have held number one since its release in Jun 2026

Lab claims count from the day the card came out, independent runs from the model’s release.

DeepSeek Anthropic
Own leaderboard 4 leaders
Claude Opus 5.5 0.21

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard Reported by the lab in its model card
0.00 2.50 5.00 7.50 10.0 2026 2027 0.00 2.50 5.00 7.50 10.0 2026 2027

Labs report it only in system cards

We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark. Some system cards do, below.

One system card reports it

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

DateDocumentScoreAs the document writes itSection
27 Jul 2026 Kimi K3 technical report 9.1 Agent Arena, rank 4 of 37 (Table 5) Other

More in Tool use

All 125 in Tool use
  • AutomationBench In use Agent completes end-to-end business workflows inside a simulated company exposing dozens of REST APIs. Graded on resulting state. Gemini 4 Carbon 53.9%
  • MCP Atlas In use Multi-step workflows over real MCP servers. Scores tool-use competency on realistic tasks. Muse Spark 1.1 88.1%
  • BFCL V4 In use Models call functions and tools, including multi-turn agentic tasks with web search and memory. Checked against expected calls and outcomes. Qwen3.7-Max 75%
  • τ³-bench Banking In use Agents act as banking customer-service reps over many turns with a simulated user, following policy. Pass^1 on the final database state. Qwen 3.8 Max 55.1%
  • Claw-Eval In use 300 human-verified agent tasks (service orchestration, multimodal, multi-turn dialogue) audited via traces and snapshots. Rubric scores with Pass^3 consistency. Claude Sonnet 4.6 81.4%
  • DeepSearchQA In use Questions that need multi-step search and synthesis. Final answers graded against references. Kimi K3 95%