How well an agent stays on the rails it was given during a session. Ranked by Arena agent_steerability, priced against live API rates. 29 models have a published score on this board.
Capability published 2026-09-08 · pricing retrieved 2026-09-10
anthropic/claude-opus-5anthropic/claude-opus-5anthropic/claude-sonnet-5anthropic/claude-opus-5Rows marked frontier are Pareto-optimal — nothing we track is both stronger and cheaper. "Strength" rescales the board so 100% = parity with the leader.
| # | Model | Provider | Score | Strength | $/1M | Context | Tools |
|---|---|---|---|---|---|---|---|
| 1 | Claude Opus 5anthropic/claude-opus-5 | Anthropic | 0.1361 | 100% | $10.00 | 1M | tools |
| 2 | Claude Fable 5anthropic/claude-fable-5 | Anthropic | 0.1100 | 81% | $20.00 | 1M | tools |
| 3 | Claude Opus 4.8anthropic/claude-opus-4.8 | Anthropic | 0.1098 | 81% | $10.00 | 1M | tools |
| 4 | Claude Sonnet 5anthropic/claude-sonnet-5 | Anthropic | 0.0974 | 72% | $4.00 | 1M | tools |
| 5 | GPT-5.6 Solopenai/gpt-5.6-sol | OpenAI | 0.0877 | 64% | $4.00 | 1.05M | tools |
| 6 | GPT-5.6 Terraopenai/gpt-5.6-terra | OpenAI | 0.0670 | 49% | $4.50 | 1.05M | tools |
| 7 | Grok 4.5x-ai/grok-4.5 | xAI | 0.0531 | 39% | $3.00 | 500K | tools |
| 8 | GLM 5.2z-ai/glm-5.2 | Z.ai (Zhipu) | 0.0292 | 21% | $1.49 | 1.048576M | tools |
| 9 | Qwen3.8 Max (0902)qwen/qwen3.8-max-0902 | Qwen (Alibaba) | 0.0276 | 20% | $3.00 | 1M | tools |
| 10 | Grok 4.6x-ai/grok-4.6 | xAI | 0.0244 | 18% | $3.00 | 500K | tools |
| 11 | GPT-5.6 Lunaopenai/gpt-5.6-luna | OpenAI | 0.0088 | 6% | $0.45 | 1.05M | tools |
| 12 | Kimi K3moonshotai/kimi-k3 | Moonshot AI | -0.0015 | -1% | $6.00 | 1.048576M | tools |
| 13 | DeepSeek V4 Pro 0813deepseek/deepseek-v4-pro-0813 | DeepSeek | -0.0092 | -7% | $1.57 | 1.048576M | tools |
| 14 | DeepSeek V4 Pro 0423deepseek/deepseek-v4-pro | DeepSeek | -0.0092 | -7% | $1.20 | 1.048576M | tools |
| 15 | GLM 5.3z-ai/glm-5.3 | Z.ai (Zhipu) | -0.0104 | -8% | $2.15 | 1.31072M | tools |
| 16 | GLM 5.3 Flashz-ai/glm-5.3-flash | Z.ai (Zhipu) | -0.0113 | -8% | $0.24 | 1.31072M | tools |
| 17 | Claude Fable 5.1anthropic/claude-fable-5.1 | Anthropic | -0.0121 | -9% | $20.00 | 1M | tools |
| 18 | DeepSeek V4 Flash 0731deepseek/deepseek-v4-flash-0731 | DeepSeek | -0.0143 | -11% | $0.10 | 1.31072M | tools |
| 19 | DeepSeek V4 Flash 0423deepseek/deepseek-v4-flash | DeepSeek | -0.0143 | -11% | $0.11 | 1.048576M | tools |
| 20 | Claude Sonnet 4.6anthropic/claude-sonnet-4.6 | Anthropic | -0.0249 | -18% | $6.00 | 1M | tools |
| 21 | Mistral Medium 3.5mistralai/mistral-medium-3-5 | Mistral | -0.0320 | -24% | $3.00 | 262K | tools |
| 22 | Gemini 3.8 Flashgoogle/gemini-3.8-flash | -0.0334 | -25% | $1.50 | 1.048576M | tools | |
| 23 | Qwen3.7 Maxqwen/qwen3.7-max | Qwen (Alibaba) | -0.0354 | -26% | $2.21 | 1M | tools |
| 24 | Qwen3.7 Plusqwen/qwen3.7-plus | Qwen (Alibaba) | -0.0381 | -28% | $0.56 | 1M | tools |
| 25 | Gemini 3.6 Flashgoogle/gemini-3.6-flash | -0.0453 | -33% | $1.50 | 1.048576M | tools |
Capability: Arena leaderboard dataset (CC-BY-4.0), Arena agent_steerability, published 2026-09-08. Pricing: OpenRouter, retrieved 2026-09-10. (input x 3 + output x 1) / 4.
curl -s https://agentleaderboards.com/api/v1/route/agent-steerability.json | jq .picksNo key, open CORS, refreshed daily. Set your own budget and constraints →
Writing & editing code · Autonomous agent work · Tool-call reliability · Recovering from shell errors · Hard reasoning · Mathematics · Precise instruction following · Long multi-turn chat · Long inputs · Creative writing · General assistant · Software & IT services · Legal & government · Medicine & healthcare · Science · Business & finance · Writing & language