Inference
EN
LMArena

Agent Arena steerability

Estimated effect of the model on whether an agent acts on a correction the user gives it.

ModelsModels with a result.54
Top resultBest result on this benchmark.33.0Claude Fable 5.2
Top-3 spreadPoints from first to third.14.4
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 effort33.0
02Gemini 4 Carbonhigh effort19.1
03Claude Opus 5.5max effort18.6
04GPT-6.1 Terramax effort16.6
05GPT-6 Astramax effort14.8
05Claude Fable 5.1max effort14.8
07Claude Sonnet 5.5max effort14.5
08GPT-6.1 Solmax effort13.4
09Gemini 4 Argonhigh effort13.2
10Muse Embermax effort12.0
11GPT-6 Solmax effort10.1
12Muse Wonder Fullmax effort9.6
13Inkling 2 Small9.3
14Hy48.8
15Claude Fable 5high effort7.9
16GPT-5.6 Solxhigh effort5.5
17Claude Opus 5max effort4.4
18DeepSeek V4.1 Flash 1017max effort4.0
19GPT-6 Lunamax effort3.5
20MiMo-V2.6-Flash2.8
20Claude Sonnet 5high effort2.8
22Muse Spark 1.3max effort2.6
22Muse Wondermax effort2.6
24Claude Haiku 5.5high effort2.4
25MiMo-V2.6-Pro1.5
26Hy4 preview0.2
27GPT-5.6 Terraxhigh effort−0.5
28Qwen3.8 Max−0.8
28Kimi K3max effort−0.8
30Step 5 Preview−0.9
31GLM-5.2max effort−1.0
32GLM-5.3max effort−1.2
33Gemini 3.8 Flashhigh effort−1.3
34Grok 4.6xhigh effort−1.8
35Grok 4.7xhigh effort−2.3
36Qwen3.8-Flash-Next−2.6
36MiniMax M3.1 Flash−2.6
38Hy3−3.1
39GLM-5.3-Flash−5.0
40Qwen3.8-27B−6.0
41Mistral Large 4−6.4
42Gemini 3.7 Flashhigh effort−7.3
43Gemini 3.1 Pro Preview−8.2
44MiniMax M3−10.4
45Qwen3.7 Plus−11.6
46Spark X2.6−12.3
47Muse Spark 1.2xhigh effort−12.5
48Qwen3.7 Max−12.8
49Muse Spark 1.1−15.4
50Mistral Medium 3.5−16.9
51NVIDIA-Nemotron-3.5-Super-VL-120B-A12B (NVFP4)−17.0
52Solar Pro 4−20.8
53Inkling−22.3
54Inkling-Small−27.7