Inference
EN
LMArena

Agent Arena task outcome

Estimated effect of the model on whether a user marks an agent session as finished.

ModelsModels with a result.54
Top resultBest result on this benchmark.22.0Claude Fable 5.2
Top-3 spreadPoints from first to third.2.5
UpdatedDate the source changed the results.11 Nov 2026

Result and price

49 of 54 results

Results

54 results
#ModelResultThe score from the source.
01Claude Fable 5.2max effort22.0
02Claude Opus 5.5max effort20.7
03Claude Sonnet 5.5max effort19.5
04Gemini 4 Carbonhigh effort18.4
05Claude Fable 5.1max effort14.4
06GPT-6 Astramax effort14.0
07Gemini 4 Argonhigh effort13.2
08GPT-6.1 Solmax effort11.7
09Muse Wonder Fullmax effort11.1
10Claude Opus 5max effort10.2
11GPT-6.1 Terramax effort9.7
12Hy46.7
13Muse Wondermax effort5.4
14Claude Fable 5high effort5.1
14Grok 4.7xhigh effort5.1
16MiMo-V2.6-Pro5.0
17GPT-6 Solmax effort4.5
17Qwen3.8 Max4.5
19MiMo-V2.6-Flash4.2
20Muse Spark 1.3max effort3.7
21Hy4 preview3.5
22Kimi K3max effort3.4
23Inkling 2 Small3.3
24GLM-5.3max effort3.1
25DeepSeek V4.1 Flash 1017max effort2.3
25GPT-5.6 Solxhigh effort2.3
27GLM-5.2max effort1.7
28Claude Haiku 5.5high effort1.6
29GLM-5.3-Flash1.3
29Gemini 3.8 Flashhigh effort1.3
31Qwen3.8-Flash-Next0.3
32GPT-6 Lunamax effort−0.4
33Claude Sonnet 5high effort−0.6
34MiniMax M3.1 Flash−0.7
35Step 5 Preview−1.5
36Grok 4.6xhigh effort−2.6
37Qwen3.8-27B−2.7
38Gemini 3.7 Flashhigh effort−3.5
39Mistral Large 4−4.7
40Muse Spark 1.1−6.5
41GPT-5.6 Terraxhigh effort−7.2
42Muse Embermax effort−7.6
43Muse Spark 1.2xhigh effort−9.8
44Spark X2.6−10.2
45Qwen3.7 Max−10.5
46Qwen3.7 Plus−10.8
47Gemini 3.1 Pro Preview−11.3
48Hy3−11.8
49Mistral Medium 3.5−16.8
50MiniMax M3−18.0
51NVIDIA-Nemotron-3.5-Super-VL-120B-A12B (NVFP4)−18.9
52Inkling−26.8
53Inkling-Small−27.6
54Solar Pro 4−29.9