---
title: "Agents · steerability: model rankings and source evidence · Undominated.ai"
canonical: https://undominated.ai/rankings/arena-agent-steerability/
description: "Agents · steerability ranked by IPS score. Inspect exact model variants, source dates, measurement limits and downloadable evidence."
---

# Agents · steerability: model rankings and source evidence · Undominated.ai

> Agents · steerability ranked by IPS score. Inspect exact model variants, source dates, measurement limits and downloadable evidence.

[All rankings](/rankings/) / Coding and agents

# Agents · steerability

Agent Arena signal scores from the official dataset. Each signal is a separate measurement; it is not a success percentage.

**58 entries** · IPS score · Higher is better

Source publication: 2026-11-04. Retrieved 2026-11-05.

Source: [Arena (lmarena-ai)](https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset) · [CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/).

Ranks use confidence bounds. Several models can share a rank; the rows within a shared rank are displayed by score.

Agents · steerability · IPS score

| Significance rank | Model / evaluated variant | IPS score | 95% confidence interval | Observations |
| --- | --- | --- | --- | --- |
| 1 | Gemini 4 Carbon (High) google | 0.193193 | 0.186058 – 0.200328 | 208,968 |
| 1 | Claude Opus 5.5 (Max) anthropic | 0.184283 | 0.164043 – 0.204523 | 36,292 |
| 2 | Claude Fable 5.1 (Max) anthropic | 0.153156 | 0.134094 – 0.172218 | 73,015 |
| 2 | GPT-6.1 Terra (Max) openai | 0.151824 | 0.137996 – 0.165653 | 118,771 |
| 3 | GPT-6 Astra (Max) openai | 0.143337 | 0.124192 – 0.162481 | 47,375 |
| 3 | Gemini 4 Argon (High) google | 0.135183 | 0.125582 – 0.144784 | 262,052 |
| 3 | GPT-6.1 Sol (Max) openai | 0.132892 | 0.121901 – 0.143882 | 96,131 |
| 3 | Claude Sonnet 5.5 (Max) anthropic | 0.123993 | 0.094852 – 0.153134 | 17,559 |
| 8 | Muse Wonder Full (Max) meta | 0.106508 | 0.095924 – 0.117091 | 186,836 |
| 8 | GPT-6 Sol (Max) openai | 0.099833 | 0.08273 – 0.116937 | 67,078 |
| 8 | Claude Fable 5 (High) anthropic | 0.08792 | 0.074152 – 0.101689 | 167,982 |
| 10 | Hy4 tencent | 0.066052 | 0.045166 – 0.086938 | 35,444 |
| 11 | Kimi K3.1 Pro (Max) moonshot | 0.065512 | 0.049588 – 0.081435 | 136,888 |
| 11 | GPT-5.6 Sol (xHigh) openai | 0.063416 | 0.049535 – 0.077296 | 166,297 |
| 12 | Claude Opus 5 (Max) anthropic | 0.053155 | 0.037246 – 0.069064 | 104,038 |
| 14 | GPT-6 Luna (Max) openai | 0.041943 | 0.036211 – 0.047674 | 1,200,732 |
| 12 | Claude Sonnet 5 (High) anthropic | 0.040254 | 0.023141 – 0.057368 | 107,206 |
| 15 | DeepSeek-V4.1-Flash-20261017 (Max) deepseek | 0.034244 | 0.026822 – 0.041665 | 463,497 |
| 15 | Muse Spark 1.3 (Max) meta | 0.033989 | 0.022938 – 0.045039 | 473,413 |
| 14 | Claude Haiku 5.5 (High) anthropic | 0.033142 | 0.020522 – 0.045761 | 271,177 |
| 14 | Muse Wonder (Max) meta | 0.03122 | 0.013051 – 0.04939 | 109,293 |
| 17 | MiMo-V2.6-Flash xiaomi | 0.021373 | 0.007531 – 0.035216 | 125,021 |
| 20 | MiMo-V2.6-Pro xiaomi | 0.010054 | -0.002537 – 0.022645 | 209,174 |
| 21 | GPT-5.6 Terra (xHigh) openai | 0.008166 | -0.002695 – 0.019027 | 223,809 |
| 22 | GPT-6 Luna (Medium) openai | 0.003787 | -0.003537 – 0.01111 | 469,939 |
| 22 | Kimi K3 (Max) moonshot | 0.001345 | -0.007038 – 0.009729 | 551,540 |
| 22 | Hy4 preview tencent | 0.000677 | -0.009153 – 0.010508 | 362,751 |
| 21 | Grok 4.6 (xHigh) xai | 0.000129 | -0.016779 – 0.017037 | 120,399 |
| 22 | Gemini 3.8 Flash (High) google | -0.002402 | -0.014105 – 0.009301 | 267,511 |
| 23 | Qwen3.8-Max alibaba | -0.0025 | -0.012048 – 0.007049 | 375,140 |
| 23 | GLM-5.2 (Max) zai | -0.006041 | -0.017911 – 0.005829 | 287,454 |
| 23 | GLM-5.3 (Max) zai | -0.008613 | -0.019923 – 0.002696 | 489,052 |
| 22 | Step 5 Preview stepfun | -0.009274 | -0.028186 – 0.009638 | 67,903 |
| 21 | DeepSeek-V4-Pro (High, 20260813) deepseek | -0.010767 | -0.034778 – 0.013244 | 28,886 |
| 23 | Grok 4.7 (xHigh) xai | -0.017538 | -0.038141 – 0.003065 | 96,104 |
| 27 | Hy3 tencent | -0.027983 | -0.047968 – -0.007999 | 52,740 |
| 32 | GPT-6 Luna (Low) openai | -0.028594 | -0.038498 – -0.01869 | 306,594 |
| 32 | Qwen3.8-Flash-Next alibaba | -0.0296 | -0.040368 – -0.018832 | 372,894 |
| 32 | MiniMax-M3.1-Flash minimax | -0.040991 | -0.063658 – -0.018324 | 43,964 |
| 34 | GLM-5.3-Flash zai | -0.042344 | -0.053246 – -0.031442 | 682,288 |
| 35 | Qwen3.8-27B alibaba | -0.050645 | -0.064499 – -0.036791 | 262,266 |
| 33 | Mistral Large 4 mistral | -0.053439 | -0.084036 – -0.022842 | 20,703 |
| 38 | Gemini 3.7 Flash (High) google | -0.05688 | -0.070904 – -0.042857 | 297,633 |
| 38 | Gemini 3.1 Pro Preview google | -0.073129 | -0.100267 – -0.045991 | 102,576 |
| 38 | Beam 1.0 501B A23B reflection | -0.082169 | -0.118022 – -0.046317 | 44,666 |
| 44 | MiniMax-M3 minimax | -0.106421 | -0.126475 – -0.086367 | 86,817 |
| 44 | Spark-X2.6 synlan | -0.109344 | -0.131329 – -0.087359 | 44,682 |
| 44 | Qwen3.7-Max alibaba | -0.110615 | -0.132567 – -0.088663 | 102,402 |
| 44 | Qwen3.7-Plus alibaba | -0.112893 | -0.139593 – -0.086193 | 47,955 |
| 44 | Muse Spark 1.2 (xHigh) meta | -0.114659 | -0.129305 – -0.100013 | 299,841 |
| 47 | Muse Spark 1.1 meta | -0.141376 | -0.154434 – -0.128318 | 473,139 |
| 46 | MiMo-V2.5-Pro xiaomi | -0.143846 | -0.164302 – -0.123391 | 101,141 |
| 49 | Mistral Medium 3.5 mistral | -0.162347 | -0.19221 – -0.132485 | 38,186 |
| 50 | Solar Pro 4 upstage | -0.170117 | -0.205238 – -0.134997 | 33,962 |
| 52 | NVIDIA-Nemotron-3.5-Super-VL-120B-A12B (NVFP4) nvidia | -0.171352 | -0.185482 – -0.157222 | 117,690 |
| 46 | Nemotron 3 Ultra nvidia | -0.183118 | -0.243841 – -0.122395 | 7,695 |
| 53 | Inkling thinky | -0.205075 | -0.238765 – -0.171385 | 50,902 |
| 56 | Inkling-Small thinky | -0.257505 | -0.293946 – -0.221064 | 38,000 |

## What this ranking can tell you

 - The population is the models in this source board, including variants and models without an available API price.
- Ranks use the published confidence bounds: one plus the number of models whose lower bound exceeds this model’s upper bound. Shared ranks are not proof of equal performance.

## Check the source

The original evidence is retained byte for byte. The downloadable board includes the source URLs, revision and SHA-256 hashes.

[Download this board and provenance →](/data/rankings/arena-agent-steerability.json)

 Source files and licences - [metadata · a7a159f557a1](/data/rankings/evidence/a7a159f557a1e594a5d41218fe8d53c63c63daa39b655e51d6fc5b482a17d059.json) · [upstream source ↗](https://huggingface.co/api/datasets/lmarena-ai/leaderboard-dataset)
- [licence · 88a46d472f35](/data/rankings/evidence/88a46d472f35274fb6b7a852c3d16c8ee8007d51b37f290161537bec26895ee4.txt) · [upstream source ↗](https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset/raw/8b6deb69ace01ee8dcda9e46eba7a5e8a808ec53/README.md)
- [rows · 8eaf40bc3f60](/data/rankings/evidence/8eaf40bc3f60879f59b202aa8ac56a6e58716aa5110f48a1798ce43c63febf38.parquet) · [upstream source ↗](https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset/resolve/8b6deb69ace01ee8dcda9e46eba7a5e8a808ec53/agent_steerability/latest-00000-of-00001.parquet)

[Explore other model types →](/rankings/)[Read the benchmark guide →](/benchmarks/)

## Continue your investigation

 - [Understand the measurements](/benchmarks/)
- [Compare task performance](/best/)
- [Compare embedding models](/embeddings/)
- [Inspect ranking rules](/methodology/)