Benchmarks / Arena (formerly LMArena)

Reported by Arena (formerly LMArena)

Arena (formerly LMArena)

Estimated effect on the agent taking a user's correction.

Last updated 2 Oct 2026

Results dated
2 Oct 2026
Results
50 configurations of 48 models
Unit
IPS effect estimate
Licence
Creative Commons Attribution 4.0 International

Arena (formerly LMArena) in brief

What it measures. Arena (formerly LMArena) is a third-party benchmark, shown here with credit. Measured here: Estimated effect on the agent taking a user's correction.

Latest results. , 48 models ranked.

Top 3 on Steerability (IPS effect estimate, higher is better, each model at its best setting):

  1. Gemini 4 Argon (high reasoning) (Google): 0.13
  2. GPT-6 Sol (max reasoning) (OpenAI): 0.11
  3. GPT-6.1 Sol (max reasoning) (OpenAI): 0.11

The first 4 are too close to tell apart, so read the order among them as a tie.

How to cite. Arena (formerly LMArena), via Spring Prompt, results dated 2 October 2026. https://springprompt.com/benchmarks/lmarena These results belong to Arena (formerly LMArena), not to us. Licence: Creative Commons Attribution 4.0 International. Credit Arena (formerly LMArena) first.

Steerability: Muse Spark 1.1

Top 15 of 48 results · IPS effect estimate, higher is better · lines show the 95% range · ≈ cannot be told apart from the leader. Choose a model to highlight it.Clear highlight

  1. 1≈ Gemini 4 Argon (high reasoning)Google 0.13
  2. 2≈ GPT-6 Sol (max reasoning)OpenAI 0.11
  3. 2≈ GPT-6.1 Sol (max reasoning)OpenAI 0.11
  4. 4≈ Claude Opus 5.5 (high reasoning)Anthropic 0.10
  5. 5 Claude Fable 5 (high reasoning)Anthropic 0.09
  6. 5 Claude Fable 5.1 (max reasoning)Anthropic 0.09
  7. 7 GPT-6 Astra (max reasoning)OpenAI 0.08
  8. 8 Claude Sonnet 5.5 (max reasoning)Anthropic 0.07
  9. 8 Claude Opus 5 (high reasoning)Anthropic 0.07
  10. 8 Claude Opus 4.8 (high reasoning)Anthropic 0.07
  11. 11 GPT-5.6 Sol (extra-high reasoning)OpenAI 0.06
  12. 12 Muse Spark 1.3 (max reasoning)Meta 0.05
  13. 12 GPT-5.4 (high reasoning)OpenAI 0.05
  14. 12 Claude Sonnet 5 (high reasoning)Anthropic 0.05
  15. 15 GPT-5.5 (extra-high reasoning)OpenAI 0.04
  16. 39 Muse Spark 1.1Meta -0.06

Full results

Arena Agent: steerability, IPS effect estimate, higher is better
#ModelSteerability · 95% range
IPS effect estimate, higher is better
Price
$ per million tokens, in / out
1≈ Gemini 4 Argon (high reasoning)Google
0.13
0.12–0.15
$2 / $10
2≈ GPT-6 Sol (max reasoning)OpenAI
0.11
0.09–0.13
$2 / $10
2≈ GPT-6.1 Sol (max reasoning)OpenAI
0.11
0.08–0.14
$2 / $10
4≈ Claude Opus 5.5 (high reasoning)Anthropic
0.10
0.08–0.13
$4 / $20
5 Claude Fable 5 (high reasoning)Anthropic
0.09
0.08–0.10
$10 / $50
5 Claude Fable 5.1 (max reasoning)Anthropic
0.09
0.06–0.12
$10 / $50
7 GPT-6 Astra (max reasoning)OpenAI
0.08
0.05–0.10
$10 / $50
8 Claude Sonnet 5.5 (max reasoning)Anthropic
0.07
0.04–0.10
$2 / $10
8 Claude Opus 5 (high reasoning)Anthropic · best of 2 settings
0.07
0.05–0.08
$5 / $25
8 Claude Opus 4.8 (high reasoning)Anthropic
0.07
0.05–0.08
$5 / $25
11 GPT-5.6 Sol (extra-high reasoning)OpenAI
0.06
0.04–0.08
$4 / $20
12 Muse Spark 1.3 (max reasoning)Meta
0.05
0.05–0.06
$1.25 / $4.25
12 GPT-5.4 (high reasoning)OpenAI
0.05
0.04–0.06
$2.50 / $15
12 Claude Sonnet 5 (high reasoning)Anthropic
0.05
0.03–0.07
$2 / $10
15 GPT-5.5 (extra-high reasoning)OpenAI · best of 2 settings
0.04
0.03–0.06
$5 / $30
15 GPT-6 Luna (max reasoning)OpenAI
0.04
0.03–0.05
$0.10 / $0.50
17 Gemini 3.8 Flash (high reasoning)Google
0.03
0.02–0.04
$1.50 / $7.50
17 Qwen3.8-Max (0902)Alibaba
0.03
0.02–0.04
$2 / $6
19 GPT-5.6 Terra (extra-high reasoning)OpenAI
0.02
0.01–0.04
$2 / $12
19 Step 5 PreviewStepFun
0.02
-0.00–0.04
$1 / $2.70
19 Grok 4.5xAI
0.02
0.01–0.03
$2 / $6
22 GLM-5.2 (max reasoning)Z.ai
0.01
0.01–0.02
$1.40 / $4.40
22 Kimi K3 (max reasoning)Moonshot AI
0.01
0.01–0.02
$3 / $15
22 DeepSeek-V4.1-Flash (max reasoning)DeepSeek
0.01
0.01–0.02
$0.30 / $1.20
22 DeepSeek-V4-Pro (0813, high reasoning)DeepSeek
0.01
-0.01–0.03
$1.32 / $3.96
22 GLM-5.3 (max reasoning)Z.ai
0.01
-0.00–0.02
$1.40 / $4.40
22 Hy4 PreviewTencent
0.01
-0.00–0.02
$0.83 / $2.50
22 Hy3Tencent
0.01
-0.01–0.02
$0.14 / $0.58
22 Grok 4.6 (extra-high reasoning)xAI
0.01
-0.01–0.02
$2 / $6
30 Grok 4.7 (extra-high reasoning)xAI
0.00
-0.02–0.03
$2 / $6
30 GPT-5.6 Luna (extra-high reasoning)OpenAI
0.00
-0.01–0.01
$0.20 / $1.20
30 MiMo-V2.6-FlashXiaomi
-0.00
-0.02–0.01
$0.14 / $0.28
30 Qwen3.8-27BAlibaba
-0.00
-0.01–0.00
$0.50 / $3
34 GLM-5.3-FlashZ.ai
-0.01
-0.01–0.00
$0.15 / $0.50
34 Qwen3.8-Flash-NextAlibaba
-0.01
-0.02–0.00
–
34 Gemini 3.7 Flash (high reasoning)Google
-0.01
-0.02–-0.00
$1.50 / $7.50
37 Gemini 3.1 Pro PreviewGoogle
-0.05
-0.06–-0.03
$2 / $12
37 Muse Spark 1.2 (extra-high reasoning)Meta
-0.05
-0.06–-0.04
$1.25 / $4.25
39 Mistral Medium 3.5Mistral AI
-0.06
-0.09–-0.04
$1.50 / $7.50
39 Muse Spark 1.1Meta
-0.06
-0.07–-0.06
$1.25 / $4.25
39 Qwen3.7-MaxAlibaba
-0.06
-0.08–-0.05
$1.48 / $4.42
42 MiniMax-M3MiniMax
-0.07
-0.09–-0.06
$0.30 / $1.20
42 Gemini 3.6 Flash (high reasoning)Google
-0.07
-0.09–-0.06
$1.50 / $7.50
44 MiMo-V2.5-ProXiaomi
-0.08
-0.09–-0.06
$0.43 / $0.87
44 Qwen3.7-PlusAlibaba
-0.08
-0.10–-0.06
$0.32 / $1.28
46 Solar Pro 4Upstage
-0.09
-0.12–-0.06
$0.15 / $0.60
47 InklingThinking Machines
-0.12
-0.14–-0.10
$1 / $4.05
48 Inkling SmallThinking Machines
-0.14
-0.17–-0.11
$0.45 / $1.20

Swipe the table sideways for more columns.

Source: Arena (formerly LMArena), via Spring Prompt, springprompt.com/benchmarks/lmarena. Test your own prompt on these models →

Ranks follow the score as shown, so equal numbers share a rank. ≈ marks results whose 95% range overlaps the leader's: they cannot be told apart from it. Each model is shown at its best setting; show every setting. Results as published by Arena (formerly LMArena); we do not re-run them.

Progress over time

Steerability, plotted at each model's release date, higher is better. Each model counts once, at its best setting. Scores are today's, not launch day's. Hover or tap a mark for its model.

Best score so far

Every model at its release date. The line moves each time a new release beats every earlier one, and each new best is named where there is room.

-0.2-0.100.10.2 Apr 2026JulOct Gemini 3.1 Pro PreviewGPT-5.4Claude Opus 4.8Claude Fable 5GPT-6 SolGemini 4 Argon

Best score by lab

Each lab's best so far, stepping whenever one of its releases beats its own earlier models. The 6 labs with the best scores today; 10 others not drawn.

GoogleOpenAIAnthropicMetaAlibabaStepFun

-0.1-0.0500.050.10.15 Apr 2026JulOct GoogleOpenAIAnthropicMetaAlibabaStepFun

What it measures

Estimated effect on the agent taking a user's correction.

What it does not measure

Not a success rate: the estimated effect of choosing this model as the agent's orchestrator, against an even mix of the models tested, inside Arena's own harness.

Contains data from the Arena Leaderboard Dataset by Arena, licensed under CC BY 4.0. Licence: Creative Commons Attribution 4.0 International.