Arena ranking
Browse Agent Steerability results, model ratings and supporting selection context.
Data as of 2026-11-07
This source leaderboard ranking orders models by their original Arena rank rather than an aggregated benchmark score.
Rank | Model | Source Agent score | LLMBoard Score | Observations | Official input / 1M | Output speed |
|---|
| Rank01 | ModelGO | Source Agent score19.32% | LLMBoard ScoreN/A | Observations209K | Official input / 1MN/A | Output speedN/A |
| Rank02 | ModelAN | Source Agent score15.32% | LLMBoard ScoreN/A | Observations73K | Official input / 1M$10 | Output speedN/A |
| Rank03 | ModelOP | Source Agent score14.33% | LLMBoard ScoreN/A | Observations47.4K | Official input / 1M$10 | Output speed48.60 tok/s |
| Rank04 | ModelGO | Source Agent score13.52% | LLMBoard ScoreN/A | Observations262.1K | Official input / 1M$2 | Output speedN/A |
| Rank05 | ModelAN | Source Agent score8.79% | LLMBoard ScoreN/A | Observations168K | Official input / 1M$10 | Output speedN/A |
| Rank06 | ModelTE | Source Agent score6.61% | LLMBoard ScoreN/A | Observations35.4K | Official input / 1MN/A | Output speedN/A |
| Rank07 | ModelOP | Source Agent score6.34% | LLMBoard ScoreN/A | Observations166.3K | Official input / 1M$4 | Output speedN/A |
| Rank08 | ModelAN | Source Agent score5.32% | LLMBoard ScoreN/A | Observations104K | Official input / 1M$5 | Output speedN/A |
| Rank09 | ModelAN | Source Agent score4.03% | LLMBoard ScoreN/A | Observations107.2K | Official input / 1M$2 | Output speed42.00 tok/s |
| Rank10 | ModelME | Source Agent score3.40% | LLMBoard ScoreN/A | Observations473.4K | Official input / 1M$1.25 | Output speed3.62 tok/s |
| Rank11 | ModelAN | Source Agent score3.31% | LLMBoard ScoreN/A | Observations271.2K | Official input / 1M$0.10 | Output speedN/A |
| Rank12 | ModelXI | Source Agent score2.14% | LLMBoard ScoreN/A | Observations125K | Official input / 1M$0.14 | Output speedN/A |
| Rank13 | ModelXI | Source Agent score1.01% | LLMBoard ScoreN/A | Observations209.2K | Official input / 1M$0.435 | Output speed4.94 tok/s |
| Rank14 | ModelOP | Source Agent score0.82% | LLMBoard ScoreN/A | Observations223.8K | Official input / 1M$2 | Output speedN/A |
| Rank15 | ModelOP | Source Agent score0.38% | LLMBoard ScoreN/A | Observations469.9K | Official input / 1M$0.10 | Output speedN/A |
| Rank16 | ModelMA | Source Agent score0.13% | LLMBoard ScoreN/A | Observations551.5K | Official input / 1M$3 | Output speedN/A |
| Rank17 | ModelXA | Source Agent score0.01% | LLMBoard ScoreN/A | Observations120.4K | Official input / 1M$2 | Output speedN/A |
| Rank18 | ModelGO | Source Agent score-0.24% | LLMBoard ScoreN/A | Observations267.5K | Official input / 1M$0.75 | Output speed107.73 tok/s |
| Rank19 | ModelAC | Source Agent score-0.25% | LLMBoard ScoreN/A | Observations375.1K | Official input / 1M$2 | Output speedN/A |
| Rank20 | ModelZA | Source Agent score-0.60% | LLMBoard ScoreN/A | Observations287.5K | Official input / 1M$1.2 | Output speedN/A |
| Rank21 | ModelZA | Source Agent score-0.86% | LLMBoard ScoreN/A | Observations489.1K | Official input / 1M$1.2 | Output speedN/A |
| Rank22 | ModelDE | Source Agent score-1.08% | LLMBoard ScoreN/A | Observations28.9K | Official input / 1M$0.66 | Output speedN/A |
| Rank23 | ModelXA | Source Agent score-1.75% | LLMBoard ScoreN/A | Observations96.1K | Official input / 1M$2 | Output speedN/A |
| Rank24 | ModelTE | Source Agent score-2.80% | LLMBoard ScoreN/A | Observations52.7K | Official input / 1MN/A | Output speedN/A |
| Rank25 | ModelAC | Source Agent score-2.96% | LLMBoard ScoreN/A | Observations372.9K | Official input / 1MN/A | Output speedN/A |
| Rank26 | ModelMI | Source Agent score-4.10% | LLMBoard ScoreN/A | Observations44K | Official input / 1M$0.15 | Output speedN/A |
| Rank27 | ModelZA | Source Agent score-4.23% | LLMBoard ScoreN/A | Observations682.3K | Official input / 1M$0.1195 | Output speed31.51 tok/s |
| Rank28 | ModelAC | Source Agent score-5.06% | LLMBoard ScoreN/A | Observations262.3K | Official input / 1MN/A | Output speedN/A |
| Rank29 | ModelMA | Source Agent score-5.34% | LLMBoard ScoreN/A | Observations20.7K | Official input / 1M$0.70 | Output speed50.35 tok/s |
| Rank30 | ModelGO | Source Agent score-5.69% | LLMBoard ScoreN/A | Observations297.6K | Official input / 1M$0.75 | Output speedN/A |
This leaderboard overview compares the source ranking, score gaps and observed task volume.
Use confidence and observed task volume to judge how firmly each leaderboard position is supported; these source proportions remain separate from benchmark coverage.
Vendor representation among the top-ranked models in this leaderboard.
A practical summary of the first five entries in this agent steerability AI model leaderboard, with benchmark evidence and pricing kept in context.
Common questions about the Agent Steerability Leaderboard.
Gemini 4 Carbon is currently ranked first with 19.32 Agent score.
The current leaders are Gemini 4 Carbon (rank #1), Claude Fable 5.1 (rank #2), and GPT-6-Astra (rank #3).
Claude Haiku 5.5 has the lowest matched official input price at $0.1 per 1M tokens.
The fastest matched records are MiniMax M3 (156.78 tok/s via MiniMax), Gemini 3.8 Flash (107.73 tok/s via Google), and Gemini 3.1 Pro (90.00 tok/s via Google).
No. Source Arena results describe a particular task or preference signal. Capability benchmarks, prices and runtime can produce different rankings.
This source-native ranking follows the original source metric. Benchmark evidence is shown separately and is not substituted for the source result.
This page currently compares 44 models.
A model may not yet have a matching Arena result, or multiple version results may resolve to the same model.
Ranking basisThis agent steerability AI model leaderboard uses the source agent score and rank. The leaderboard ranking keeps matched price and speed data separate from benchmark evidence.
Selection summary
Gemini 4 Carbon currently leads the agent steerability ranking at 19.32%. Compare evidence coverage, source activity and price separately before choosing a model for production.
Use this leaderboard with the supporting benchmark results and coverage details above. A leaderboard position summarizes the selected ranking signal; it does not replace workload-specific testing.