Inference
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.43
Top resultBest result on this test.11.1Claude Opus 4.8
Top-3 spreadPoints from first to third.2.7
UpdatedDate the source changed the results.15 Sept 2026

Result and price

-15-10-5051015$0.3$1$3$10$30$100RESULTOUTPUT PRICE PER 1M TOKENS · LOG SCALE

Results

43 results
#ModelResultThe score from the source.
01Claude Opus 4.8high effort11.1
02Claude Fable 5high effort10.6
03Claude Sonnet 5high effort8.4
04Claude Opus 5max effort6.8
05GPT-5.6 Solxhigh effort6.6
06GPT-5.5xhigh effort6.6
07Grok 4.55.9
08GPT-5.6 Terraxhigh effort4.9
09GLM-5.2max effort4.9
10Grok 4.6xhigh effort4.7
11Claude Fable 5.1max effort3.9
12Qwen3.8max effort3.6
13GPT-5.4high effort2.5
14GPT-5.6 Lunaxhigh effort2.1
15GLM-5.3max effort1.9
16Muse Spark 1.3max effort1.5
17DeepSeek V4 Pro (High) (0813)1.4
18Deepseek V4 Flash (High) (20260731)0.4
19GLM-5.3-Flash0.1
20Qwen3.8-27B0.1
21Gemini 3.7 Flashhigh effort-0.3
22GPT-6 Astramax effort-0.5
23Gemini 3.8 Flashhigh effort-1.2
24Hy4 preview-1.2
25DeepSeek V4.1 Flashmax effort-1.6
26Hy3-1.8
27DeepSeek V4 Pro 0423-2.1
28Qwen3.7max effort-2.7
29Gemini 3.6 Flashhigh effort-3.4
30Gemini 3.1 Pro Preview-3.6
31MiMo-V2.5-Pro-4.0
32Muse Spark 1.1-4.2
33Muse Spark 1.2xhigh effort-4.6
34Qwen3.8-Flash-Next-4.7
35Claude Sonnet 4.6-5.0
36Qwen3.7 Plus-5.4
37Mistral Medium 3.5-6.0
38MiniMax M3-6.6
39MiniMax M2.7-6.6
40Solar Pro 4-10.1
41Inkling-10.5
42Inkling-Small-11.8
43Gemini 3.5 Flash-Lite-12.8