InferenceBench
Coding agents optimize a language model inference server within a fixed time budget. The harness measures serving speed and checks answer quality against a reference.
The board lists separate configurations for different agents, reasoning settings and software versions.
- Made by
- ELLIS Institute Tübingen and collaborators
- Kind
- Machine learning engineering
- Level
- Expert
- Scored
- Checked automatically
- Answers
- Agent task
- Tasks
- All public
- Measures
- Efficiency
- Language
- English
- Interaction
- Agentic
- Environment
- Terminal
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Agents
- Strongest model tested
- Claude Fable 5.2 #2 on the capability index
- Last model added
- November 2026
- Scores refreshed
- November 20th, 2026
Claude Opus 5.5 leads with 12.1× Claude Opus 5.5 leads with 12.1×
Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran on its own with several settings, we keep its best. Every run is in the next chart. Agents other companies built around a model get a pale shade of that model's lab color. Gray marks blends and entries not in our model list, such as fine-tunes and older models. None of these count toward leaders. Agents built around a model, such as OpenCode + Gemini 4 Neon, are their own entries. We don't count them as the model's score. This view shows only models we track, each at its best.
- 12.1× Claude Opus 5.5
- 12.1× Claude Sonnet 5.5
- 11.0× OpenCode + Gemini 4 Neon
- 10.8× Codex CLI 0.169.0 + GPT-6.2 Astra
- 10.5× OpenCode + Gemini 4 Carbon
- 10.3× Claude Fable 5.2
- 9.83× Claude Fable 5.1
- 8.93× OpenCode + Gemini 4 Hydrogen
- 8.90× Claude Opus 5
- 8.74× Claude Fable 5
- 8.53× Claude Opus 4.7
- 7.94× Codex CLI 0.169.0 + GPT-6.2 Sol
- 7.90× GPT-6 Astra
- 7.64× GPT-6 Sol
- 7.60× Claude Opus 4.8
- 7.34× GPT-5.6 Sol
- 7.00× Claude Code + GLM-5.2
- 6.70× OpenCode + Gemini 4 Argon
- 6.65× Grok 4.6
- 6.43× Claude Sonnet 5
- 6.27× OpenCode + Kimi K2.7 Code
- 6.16× GPT-5.4
- 5.70× OpenCode + Kimi K3
- 5.59× OpenCode + Gemini 3.8 Flash
- 5.56× Claude Sonnet 4.6
- 5.53× GPT-5.3 Codex
- 5.45× GPT-5.5
- 4.98× Claude Code + GLM-5.3
- 4.92× OpenCode + Gemini 3.1 Pro
- 4.53× OpenCode + Muse Spark 1.3
- 4.51× OpenCode + Kimi K2.6
- 4.49× OpenCode + GLM-5.3-Flash
- 4.43× Claude Haiku 5.5
- 4.38× Claude Opus 4.6
- 4.28× GPT-5.2
- 4.16× OpenCode + Gemini 3.5 Flash
- 3.76× Claude Opus 4.5
- 3.70× Grok 4.5
- 3.59× GPT-5.1-Codex-Max
- 3.42× OpenCode + Grok 4.5
- 3.22× OpenCode + GLM-5
- 3.18× Claude Sonnet 4.5
- 2.78× Claude Haiku 4.5
- 1.98× GPT-5.2 Codex
- 1 Claude Opus 5.5 Max, Claude Code · v2.1.280 · prompt v1.0.1 12.1×
- 2 Claude Sonnet 5.5 Max, Claude Code · v2.1.280 · prompt v1.0.1 12.1×
- 3 OpenCode + Gemini 4 Neon High, OpenCode · prompt v1.0.1 11.0×
- 4 Codex CLI 0.169.0 + GPT-6.2 Astra Ultra, Codex CLI 0.169.0 · prompt v1.0.1 10.8×
- 5 OpenCode + Gemini 4 Carbon High, OpenCode · prompt v1.0.1 10.5×
- 6 Claude Fable 5.2 Max, Claude Code · v2.1.321 · prompt v1.0.1 10.3×
- 7 Claude Fable 5.1 Claude Code · v2.1.258 · prompt v1.0.1 9.83×
- 8 OpenCode + Gemini 4 Hydrogen OpenCode · prompt v1.0.1 8.93×
- 9 Claude Opus 5 Claude Code · v2.1.119 · prompt v1.0.1 8.90×
- 10 Claude Fable 5 Low, Claude Code · v2.1.175 · prompt v1.0.1 8.74×
- 11 Claude Opus 4.7 Claude Code · v2.1.175 8.53×
- 12 Codex CLI 0.169.0 + GPT-6.2 Sol Ultra, Codex CLI 0.169.0 · prompt v1.0.1 7.94×
- 13 GPT-6 Astra Ultra, Codex CLI · prompt v1.0.1 7.90×
- 14 GPT-6 Sol Ultra, Codex CLI · prompt v1.0.1 7.64×
- 15 Claude Opus 4.8 Claude Code · v2.1.175 7.60×
- 16 GPT-5.6 Sol Ultra, Codex CLI · prompt v1.0.1 7.34×
- 17 Claude Code + GLM-5.2 Max, Claude Code · v2.1.119 · prompt v1.0.1 7.00×
- 18 OpenCode + Gemini 4 Argon High, OpenCode · prompt v1.0.1 6.70×
- 19 Grok 4.6 Grok Build · prompt v1.0.1 6.65×
- 20 Claude Sonnet 5 Claude Code · v2.1.119 · prompt v1.0.1 6.43×
- 21 OpenCode + Kimi K2.7 Code OpenCode · prompt v1.0.1 6.27×
- 22 GPT-5.4 High, Codex CLI 6.16×
- 23 OpenCode + Kimi K3 OpenCode · prompt v1.0.1 5.70×
- 24 OpenCode + Gemini 3.8 Flash OpenCode · prompt v1.0.1 5.59×
- 25 Claude Sonnet 4.6 Claude Code 5.56×
- 26 GPT-5.3 Codex High, Codex CLI 5.53×
- 27 GPT-5.5 xHigh, Codex CLI 5.45×
- 28 Claude Code + GLM-5.3 Max, Claude Code · v2.1.119 · prompt v1.0.1 4.98×
- 29 OpenCode + Gemini 3.1 Pro OpenCode 4.92×
- 30 OpenCode + Muse Spark 1.3 OpenCode · prompt v1.0.1 4.53×
- 31 OpenCode + Kimi K2.6 OpenCode 4.51×
- 32 OpenCode + GLM-5.3-Flash OpenCode · prompt v1.0.1 4.49×
- 33 Claude Haiku 5.5 Max, Claude Code · v2.1.321 · prompt v1.0.1 4.43×
- 34 Claude Opus 4.6 Claude Code 4.38×
- 35 GPT-5.2 Codex CLI 4.28×
- 36 OpenCode + Gemini 3.5 Flash OpenCode 4.16×
- 37 Claude Opus 4.5 Claude Code 3.76×
- 38 Grok 4.5 Grok Build · prompt v1.0.1 3.70×
- 39 GPT-5.1-Codex-Max Codex CLI 3.59×
- 40 OpenCode + Grok 4.5 OpenCode · prompt v1.0.1 3.42×
- 41 OpenCode + GLM-5 OpenCode 3.22×
- 42 Claude Sonnet 4.5 Claude Code 3.18×
- 43 Claude Haiku 4.5 Claude Code 2.78×
- 44 GPT-5.2 Codex Codex CLI 1.98×
Source: Official InferenceBench leaderboard
- 12.1× Claude Opus 5.5
- 12.1× Claude Sonnet 5.5
- 10.3× Claude Fable 5.2
- 9.83× Claude Fable 5.1
- 8.90× Claude Opus 5
- 8.74× Claude Fable 5
- 8.53× Claude Opus 4.7
- 7.90× GPT-6 Astra
- 7.64× GPT-6 Sol
- 7.60× Claude Opus 4.8
- 7.34× GPT-5.6 Sol
- 6.65× Grok 4.6
- 6.43× Claude Sonnet 5
- 6.16× GPT-5.4
- 5.56× Claude Sonnet 4.6
- 5.53× GPT-5.3 Codex
- 5.45× GPT-5.5
- 4.43× Claude Haiku 5.5
- 4.38× Claude Opus 4.6
- 4.28× GPT-5.2
- 3.76× Claude Opus 4.5
- 3.70× Grok 4.5
- 3.59× GPT-5.1-Codex-Max
- 3.18× Claude Sonnet 4.5
- 2.78× Claude Haiku 4.5
- 1.98× GPT-5.2 Codex
- 1 Claude Opus 5.5 Max, Claude Code · v2.1.280 · prompt v1.0.1 12.1×
- 2 Claude Sonnet 5.5 Max, Claude Code · v2.1.280 · prompt v1.0.1 12.1×
- 3 Claude Fable 5.2 Max, Claude Code · v2.1.321 · prompt v1.0.1 10.3×
- 4 Claude Fable 5.1 Claude Code · v2.1.258 · prompt v1.0.1 9.83×
- 5 Claude Opus 5 Claude Code · v2.1.119 · prompt v1.0.1 8.90×
- 6 Claude Fable 5 Low, Claude Code · v2.1.175 · prompt v1.0.1 8.74×
- 7 Claude Opus 4.7 Claude Code · v2.1.175 8.53×
- 8 GPT-6 Astra Ultra, Codex CLI · prompt v1.0.1 7.90×
- 9 GPT-6 Sol Ultra, Codex CLI · prompt v1.0.1 7.64×
- 10 Claude Opus 4.8 Claude Code · v2.1.175 7.60×
- 11 GPT-5.6 Sol Ultra, Codex CLI · prompt v1.0.1 7.34×
- 12 Grok 4.6 Grok Build · prompt v1.0.1 6.65×
- 13 Claude Sonnet 5 Claude Code · v2.1.119 · prompt v1.0.1 6.43×
- 14 GPT-5.4 High, Codex CLI 6.16×
- 15 Claude Sonnet 4.6 Claude Code 5.56×
- 16 GPT-5.3 Codex High, Codex CLI 5.53×
- 17 GPT-5.5 xHigh, Codex CLI 5.45×
- 18 Claude Haiku 5.5 Max, Claude Code · v2.1.321 · prompt v1.0.1 4.43×
- 19 Claude Opus 4.6 Claude Code 4.38×
- 20 GPT-5.2 Codex CLI 4.28×
- 21 Claude Opus 4.5 Claude Code 3.76×
- 22 Grok 4.5 Grok Build · prompt v1.0.1 3.70×
- 23 GPT-5.1-Codex-Max Codex CLI 3.59×
- 24 Claude Sonnet 4.5 Claude Code 3.18×
- 25 Claude Haiku 4.5 Claude Code 2.78×
- 26 GPT-5.2 Codex Codex CLI 1.98×
Source: Official InferenceBench leaderboard
| # | Entry | Score ± | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 | 12.1× ±0.6 | Max, Claude Code · v2.1.280 · prompt v1.0.1, Claude Code | Sep 2026 | Official leaderboard |
| 2 | Claude Sonnet 5.5 | 12.1× ±1.4 | Max, Claude Code · v2.1.280 · prompt v1.0.1, Claude Code | Sep 2026 | Official leaderboard |
| 3 | OpenCode + Gemini 4 Neon | 11.0× ±0.9 | High, OpenCode · prompt v1.0.1 | Nov 2026 | Official leaderboard |
| 4 | Codex CLI 0.169.0 + GPT-6.2 Astra | 10.8× ±0.7 | Ultra, Codex CLI 0.169.0 · prompt v1.0.1 | Nov 2026 | Official leaderboard |
| 5 | OpenCode + Gemini 4 Carbon | 10.5× ±1.1 | High, OpenCode · prompt v1.0.1 | Oct 2026 | Official leaderboard |
| 6 | Claude Fable 5.2 | 10.3× ±1.6 | Max, Claude Code · v2.1.321 · prompt v1.0.1, Claude Code | Nov 2026 | Official leaderboard |
| 7 | Claude Fable 5.1 | 9.83× ±0.5 | Claude Code · v2.1.258 · prompt v1.0.1, Claude Code | Sep 2026 | Official leaderboard |
| 8 | OpenCode + Gemini 4 Hydrogen | 8.93× ±0.3 | OpenCode · prompt v1.0.1 | Nov 2026 | Official leaderboard |
| 9 | Claude Opus 5 | 8.90× ±1.3 | Claude Code · v2.1.119 · prompt v1.0.1, Claude Code | Jul 2026 | Official leaderboard |
| 10 | Claude Fable 5 | 8.74× ±1.9 | Low, Claude Code · v2.1.175 · prompt v1.0.1, Claude Code | Jun 2026 | Official leaderboard |
| 11 | Claude Opus 4.7 | 8.53× ±0.8 | Claude Code · v2.1.175, Claude Code | Apr 2026 | Official leaderboard |
| 12 | Codex CLI 0.169.0 + GPT-6.2 Sol | 7.94× ±0.6 | Ultra, Codex CLI 0.169.0 · prompt v1.0.1 | Nov 2026 | Official leaderboard |
| 13 | GPT-6 Astra | 7.90× ±0.3 | Ultra, Codex CLI · prompt v1.0.1, Codex CLI | Sep 2026 | Official leaderboard |
| 14 | GPT-6 Sol | 7.64× ±0.1 | Ultra, Codex CLI · prompt v1.0.1, Codex CLI | Sep 2026 | Official leaderboard |
| 15 | Claude Opus 4.8 | 7.60× ±2.2 | Claude Code · v2.1.175, Claude Code | May 2026 | Official leaderboard |
| 16 | GPT-5.6 Sol | 7.34× ±0.2 | Ultra, Codex CLI · prompt v1.0.1, Codex CLI | Jul 2026 | Official leaderboard |
| 17 | Claude Code + GLM-5.2 | 7.00× ±0.2 | Max, Claude Code · v2.1.119 · prompt v1.0.1 | Jun 2026 | Official leaderboard |
| 18 | OpenCode + Gemini 4 Argon | 6.70× ±0.3 | High, OpenCode · prompt v1.0.1 | Sep 2026 | Official leaderboard |
| 19 | Grok 4.6 | 6.65× ±1.1 | Grok Build · prompt v1.0.1, Grok Build | Aug 2026 | Official leaderboard |
| 20 | Claude Sonnet 5 | 6.43× ±1.5 | Claude Code · v2.1.119 · prompt v1.0.1, Claude Code | Jun 2026 | Official leaderboard |
| 21 | OpenCode + Kimi K2.7 Code | 6.27× ±0.2 | OpenCode · prompt v1.0.1 | Jun 2026 | Official leaderboard |
| 22 | GPT-5.4 | 6.16× ±1.2 | High, Codex CLI, Codex CLI | Mar 2026 | Official leaderboard |
| 23 | OpenCode + Kimi K3 | 5.70× ±1.7 | OpenCode · prompt v1.0.1 | Jul 2026 | Official leaderboard |
| 24 | OpenCode + Gemini 3.8 Flash | 5.59× ±0.5 | OpenCode · prompt v1.0.1 | Sep 2026 | Official leaderboard |
| 25 | Claude Sonnet 4.6 | 5.56× ±1.6 | Claude Code, Claude Code | Feb 2026 | Official leaderboard |
| 26 | GPT-5.3 Codex | 5.53× ±0.5 | High, Codex CLI, Codex CLI | Feb 2026 | Official leaderboard |
| 27 | GPT-5.5 | 5.45× ±1.3 | xHigh, Codex CLI, Codex CLI | Apr 2026 | Official leaderboard |
| 28 | Claude Code + GLM-5.3 | 4.98× ±0.9 | Max, Claude Code · v2.1.119 · prompt v1.0.1 | Aug 2026 | Official leaderboard |
| 29 | OpenCode + Gemini 3.1 Pro | 4.92× ±0.8 | OpenCode | Feb 2026 | Official leaderboard |
| 30 | OpenCode + Muse Spark 1.3 | 4.53× ±0.7 | OpenCode · prompt v1.0.1 | Sep 2026 | Official leaderboard |
| 31 | OpenCode + Kimi K2.6 | 4.51× ±0.5 | OpenCode | Apr 2026 | Official leaderboard |
| 32 | OpenCode + GLM-5.3-Flash | 4.49× ±0.7 | OpenCode · prompt v1.0.1 | Aug 2026 | Official leaderboard |
| 33 | Claude Haiku 5.5 | 4.43× ±0.8 | Max, Claude Code · v2.1.321 · prompt v1.0.1, Claude Code | Oct 2026 | Official leaderboard |
| 34 | Claude Opus 4.6 | 4.38× ±1.3 | Claude Code, Claude Code | Feb 2026 | Official leaderboard |
| 35 | GPT-5.2 | 4.28× ±1.3 | Codex CLI, Codex CLI | Dec 2025 | Official leaderboard |
| 36 | OpenCode + Gemini 3.5 Flash | 4.16× ±0.7 | OpenCode | May 2026 | Official leaderboard |
| 37 | Claude Opus 4.5 | 3.76× ±0.9 | Claude Code, Claude Code | Nov 2025 | Official leaderboard |
| 38 | Grok 4.5 | 3.70× ±0.8 | Grok Build · prompt v1.0.1, Grok Build | Jul 2026 | Official leaderboard |
| 39 | GPT-5.1-Codex-Max | 3.59× ±1.2 | Codex CLI, Codex CLI | Nov 2025 | Official leaderboard |
| 40 | OpenCode + Grok 4.5 | 3.42× ±1.2 | OpenCode · prompt v1.0.1 | Jul 2026 | Official leaderboard |
| 41 | OpenCode + GLM-5 | 3.22× ±0.9 | OpenCode | Feb 2026 | Official leaderboard |
| 42 | Claude Sonnet 4.5 | 3.18× ±0.9 | Claude Code, Claude Code | Sep 2025 | Official leaderboard |
| 43 | Claude Haiku 4.5 | 2.78× ±0.6 | Claude Code, Claude Code | Oct 2025 | Official leaderboard |
| 44 | GPT-5.2 Codex | 1.98× ±0.2 | Codex CLI, Codex CLI | Dec 2025 | Official leaderboard |
| # | Model | Score ± | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 | 12.1× ±0.6 | Max, Claude Code · v2.1.280 · prompt v1.0.1, Claude Code | Sep 2026 | Official leaderboard |
| 2 | Claude Sonnet 5.5 | 12.1× ±1.4 | Max, Claude Code · v2.1.280 · prompt v1.0.1, Claude Code | Sep 2026 | Official leaderboard |
| 3 | Claude Fable 5.2 | 10.3× ±1.6 | Max, Claude Code · v2.1.321 · prompt v1.0.1, Claude Code | Nov 2026 | Official leaderboard |
| 4 | Claude Fable 5.1 | 9.83× ±0.5 | Claude Code · v2.1.258 · prompt v1.0.1, Claude Code | Sep 2026 | Official leaderboard |
| 5 | Claude Opus 5 | 8.90× ±1.3 | Claude Code · v2.1.119 · prompt v1.0.1, Claude Code | Jul 2026 | Official leaderboard |
| 6 | Claude Fable 5 | 8.74× ±1.9 | Low, Claude Code · v2.1.175 · prompt v1.0.1, Claude Code | Jun 2026 | Official leaderboard |
| 7 | Claude Opus 4.7 | 8.53× ±0.8 | Claude Code · v2.1.175, Claude Code | Apr 2026 | Official leaderboard |
| 8 | GPT-6 Astra | 7.90× ±0.3 | Ultra, Codex CLI · prompt v1.0.1, Codex CLI | Sep 2026 | Official leaderboard |
| 9 | GPT-6 Sol | 7.64× ±0.1 | Ultra, Codex CLI · prompt v1.0.1, Codex CLI | Sep 2026 | Official leaderboard |
| 10 | Claude Opus 4.8 | 7.60× ±2.2 | Claude Code · v2.1.175, Claude Code | May 2026 | Official leaderboard |
| 11 | GPT-5.6 Sol | 7.34× ±0.2 | Ultra, Codex CLI · prompt v1.0.1, Codex CLI | Jul 2026 | Official leaderboard |
| 12 | Grok 4.6 | 6.65× ±1.1 | Grok Build · prompt v1.0.1, Grok Build | Aug 2026 | Official leaderboard |
| 13 | Claude Sonnet 5 | 6.43× ±1.5 | Claude Code · v2.1.119 · prompt v1.0.1, Claude Code | Jun 2026 | Official leaderboard |
| 14 | GPT-5.4 | 6.16× ±1.2 | High, Codex CLI, Codex CLI | Mar 2026 | Official leaderboard |
| 15 | Claude Sonnet 4.6 | 5.56× ±1.6 | Claude Code, Claude Code | Feb 2026 | Official leaderboard |
| 16 | GPT-5.3 Codex | 5.53× ±0.5 | High, Codex CLI, Codex CLI | Feb 2026 | Official leaderboard |
| 17 | GPT-5.5 | 5.45× ±1.3 | xHigh, Codex CLI, Codex CLI | Apr 2026 | Official leaderboard |
| 18 | Claude Haiku 5.5 | 4.43× ±0.8 | Max, Claude Code · v2.1.321 · prompt v1.0.1, Claude Code | Oct 2026 | Official leaderboard |
| 19 | Claude Opus 4.6 | 4.38× ±1.3 | Claude Code, Claude Code | Feb 2026 | Official leaderboard |
| 20 | GPT-5.2 | 4.28× ±1.3 | Codex CLI, Codex CLI | Dec 2025 | Official leaderboard |
| 21 | Claude Opus 4.5 | 3.76× ±0.9 | Claude Code, Claude Code | Nov 2025 | Official leaderboard |
| 22 | Grok 4.5 | 3.70× ±0.8 | Grok Build · prompt v1.0.1, Grok Build | Jul 2026 | Official leaderboard |
| 23 | GPT-5.1-Codex-Max | 3.59× ±1.2 | Codex CLI, Codex CLI | Nov 2025 | Official leaderboard |
| 24 | Claude Sonnet 4.5 | 3.18× ±0.9 | Claude Code, Claude Code | Sep 2025 | Official leaderboard |
| 25 | Claude Haiku 4.5 | 2.78× ±0.6 | Claude Code, Claude Code | Oct 2025 | Official leaderboard |
| 26 | GPT-5.2 Codex | 1.98× ±0.2 | Codex CLI, Codex CLI | Dec 2025 | Official leaderboard |
More effort adds a median 3 speedup
Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.
Source: Official InferenceBench leaderboard
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| Claude Opus 5.5 | Max, Claude Code · v2.1.280 · prompt v1.0.1 | Claude Code | 12.1× | ||
| Claude Sonnet 5.5 | Max, Claude Code · v2.1.280 · prompt v1.0.1 | Claude Code | 12.1× | ||
| OpenCode + Gemini 4 Neon | High, OpenCode · prompt v1.0.1 | OpenCode | 11.0× | ||
| Codex CLI 0.169.0 + GPT-6.2 Astra | Ultra, Codex CLI 0.169.0 · prompt v1.0.1 | Codex CLI 0.169.0 | 10.8× | ||
| OpenCode + Gemini 4 Carbon | High, OpenCode · prompt v1.0.1 | OpenCode | 10.5× | ||
| Claude Fable 5.2 | Max, Claude Code · v2.1.321 · prompt v1.0.1 | Claude Code | 10.3× | ||
| Claude Fable 5.1 | Claude Code · v2.1.258 · prompt v1.0.1 | Claude Code | 9.83× | ||
| Claude Fable 5.1 | Max, Claude Code · v2.1.258 · prompt v1.0.1 | Claude Code | 9.31× | ||
| Claude Sonnet 5.5 | Claude Code · v2.1.280 · prompt v1.0.1 | Claude Code | 9.06× | ||
| OpenCode + Gemini 4 Hydrogen | OpenCode · prompt v1.0.1 | OpenCode | 8.93× | ||
| Claude Opus 5 | Claude Code · v2.1.119 · prompt v1.0.1 | Claude Code | 8.90× | ||
| Claude Fable 5 | Low, Claude Code · v2.1.175 · prompt v1.0.1 | Claude Code | 8.74× | ||
| Claude Opus 4.7 | Claude Code · v2.1.175 | Claude Code | 8.53× | ||
| Codex CLI 0.169.0 + GPT-6.2 Sol | Ultra, Codex CLI 0.169.0 · prompt v1.0.1 | Codex CLI 0.169.0 | 7.94× | ||
| GPT-6 Astra | Ultra, Codex CLI · prompt v1.0.1 | Codex CLI | 7.90× | ||
| GPT-6 Sol | Ultra, Codex CLI · prompt v1.0.1 | Codex CLI | 7.64× | ||
| Claude Opus 4.8 | Claude Code · v2.1.175 | Claude Code | 7.60× | ||
| Claude Fable 5 | Claude Code · v2.1.175 · prompt v1.0.1 | Claude Code | 7.52× | ||
| Claude Opus 4.8 | xHigh, Claude Code | Claude Code | 7.34× | ||
| GPT-5.6 Sol | Ultra, Codex CLI · prompt v1.0.1 | Codex CLI | 7.34× | ||
| Claude Code + GLM-5.2 | Max, Claude Code · v2.1.119 · prompt v1.0.1 | Claude Code | 7.00× | ||
| OpenCode + Gemini 4 Argon | High, OpenCode · prompt v1.0.1 | OpenCode | 6.70× | ||
| Grok 4.6 | Grok Build · prompt v1.0.1 | Grok Build | 6.65× | ||
| Claude Sonnet 5 | Claude Code · v2.1.119 · prompt v1.0.1 | Claude Code | 6.43× | ||
| OpenCode + Kimi K2.7 Code | OpenCode · prompt v1.0.1 | OpenCode | 6.27× | ||
| GPT-5.4 | High, Codex CLI | Codex CLI | 6.16× | ||
| OpenCode + Kimi K3 | OpenCode · prompt v1.0.1 | OpenCode | 5.70× | ||
| OpenCode + Gemini 3.8 Flash | OpenCode · prompt v1.0.1 | OpenCode | 5.59× | ||
| Claude Sonnet 4.6 | Claude Code | Claude Code | 5.56× | ||
| GPT-5.3 Codex | High, Codex CLI | Codex CLI | 5.53× | ||
| GPT-5.5 | xHigh, Codex CLI | Codex CLI | 5.45× | ||
| Claude Code + GLM-5.3 | Max, Claude Code · v2.1.119 · prompt v1.0.1 | Claude Code | 4.98× | ||
| OpenCode + Gemini 3.1 Pro | OpenCode | OpenCode | 4.92× | ||
| OpenCode + Muse Spark 1.3 | OpenCode · prompt v1.0.1 | OpenCode | 4.53× | ||
| OpenCode + Kimi K2.6 | OpenCode | OpenCode | 4.51× | ||
| OpenCode + GLM-5.3-Flash | OpenCode · prompt v1.0.1 | OpenCode | 4.49× | ||
| Claude Haiku 5.5 | Max, Claude Code · v2.1.321 · prompt v1.0.1 | Claude Code | 4.43× | ||
| Claude Opus 4.6 | Claude Code | Claude Code | 4.38× | ||
| GPT-5.2 | Codex CLI | Codex CLI | 4.28× | ||
| GPT-5.5 | High, Codex CLI | Codex CLI | 4.22× | ||
| OpenCode + Gemini 3.5 Flash | OpenCode | OpenCode | 4.16× | ||
| Claude Opus 4.5 | Claude Code | Claude Code | 3.76× | ||
| Grok 4.5 | Grok Build · prompt v1.0.1 | Grok Build | 3.70× | ||
| GPT-5.1-Codex-Max | Codex CLI | Codex CLI | 3.59× | ||
| OpenCode + Grok 4.5 | OpenCode · prompt v1.0.1 | OpenCode | 3.42× | ||
| OpenCode + GLM-5 | OpenCode | OpenCode | 3.22× | ||
| Claude Sonnet 4.5 | Claude Code | Claude Code | 3.18× | ||
| Claude Fable 5 | Claude Code · v2.1.175 | Claude Code | 3.16× | ||
| Claude Haiku 4.5 | Claude Code | Claude Code | 2.78× | ||
| GPT-5.3 Codex | Medium, Codex CLI | Codex CLI | 2.32× | ||
| Claude Opus 4.7 | Claude Code · v2.1.114 | Claude Code | 2.25× | ||
| Claude Fable 5 | Low, Claude Code · v2.1.175 | Claude Code | 2.15× | ||
| GPT-5.2 Codex | Codex CLI | Codex CLI | 1.98× |
More effort adds a median 3 speedup
Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.
Source: Official InferenceBench leaderboard
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| Claude Opus 5.5 | Max, Claude Code · v2.1.280 · prompt v1.0.1 | Claude Code | 12.1× | ||
| Claude Sonnet 5.5 | Max, Claude Code · v2.1.280 · prompt v1.0.1 | Claude Code | 12.1× | ||
| Claude Fable 5.2 | Max, Claude Code · v2.1.321 · prompt v1.0.1 | Claude Code | 10.3× | ||
| Claude Fable 5.1 | Claude Code · v2.1.258 · prompt v1.0.1 | Claude Code | 9.83× | ||
| Claude Fable 5.1 | Max, Claude Code · v2.1.258 · prompt v1.0.1 | Claude Code | 9.31× | ||
| Claude Sonnet 5.5 | Claude Code · v2.1.280 · prompt v1.0.1 | Claude Code | 9.06× | ||
| Claude Opus 5 | Claude Code · v2.1.119 · prompt v1.0.1 | Claude Code | 8.90× | ||
| Claude Fable 5 | Low, Claude Code · v2.1.175 · prompt v1.0.1 | Claude Code | 8.74× | ||
| Claude Opus 4.7 | Claude Code · v2.1.175 | Claude Code | 8.53× | ||
| GPT-6 Astra | Ultra, Codex CLI · prompt v1.0.1 | Codex CLI | 7.90× | ||
| GPT-6 Sol | Ultra, Codex CLI · prompt v1.0.1 | Codex CLI | 7.64× | ||
| Claude Opus 4.8 | Claude Code · v2.1.175 | Claude Code | 7.60× | ||
| Claude Fable 5 | Claude Code · v2.1.175 · prompt v1.0.1 | Claude Code | 7.52× | ||
| Claude Opus 4.8 | xHigh, Claude Code | Claude Code | 7.34× | ||
| GPT-5.6 Sol | Ultra, Codex CLI · prompt v1.0.1 | Codex CLI | 7.34× | ||
| Grok 4.6 | Grok Build · prompt v1.0.1 | Grok Build | 6.65× | ||
| Claude Sonnet 5 | Claude Code · v2.1.119 · prompt v1.0.1 | Claude Code | 6.43× | ||
| GPT-5.4 | High, Codex CLI | Codex CLI | 6.16× | ||
| Claude Sonnet 4.6 | Claude Code | Claude Code | 5.56× | ||
| GPT-5.3 Codex | High, Codex CLI | Codex CLI | 5.53× | ||
| GPT-5.5 | xHigh, Codex CLI | Codex CLI | 5.45× | ||
| Claude Haiku 5.5 | Max, Claude Code · v2.1.321 · prompt v1.0.1 | Claude Code | 4.43× | ||
| Claude Opus 4.6 | Claude Code | Claude Code | 4.38× | ||
| GPT-5.2 | Codex CLI | Codex CLI | 4.28× | ||
| GPT-5.5 | High, Codex CLI | Codex CLI | 4.22× | ||
| Claude Opus 4.5 | Claude Code | Claude Code | 3.76× | ||
| Grok 4.5 | Grok Build · prompt v1.0.1 | Grok Build | 3.70× | ||
| GPT-5.1-Codex-Max | Codex CLI | Codex CLI | 3.59× | ||
| Claude Sonnet 4.5 | Claude Code | Claude Code | 3.18× | ||
| Claude Fable 5 | Claude Code · v2.1.175 | Claude Code | 3.16× | ||
| Claude Haiku 4.5 | Claude Code | Claude Code | 2.78× | ||
| GPT-5.3 Codex | Medium, Codex CLI | Codex CLI | 2.32× | ||
| Claude Opus 4.7 | Claude Code · v2.1.114 | Claude Code | 2.25× | ||
| Claude Fable 5 | Low, Claude Code · v2.1.175 | Claude Code | 2.15× | ||
| GPT-5.2 Codex | Codex CLI | Codex CLI | 1.98× |
12 models have held number one on the official leaderboard since 2023
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official InferenceBench leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official InferenceBench leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in AI research
- PaperBench In use Agent replicates ICML 2024 papers from scratch. LLM judge grades against author-written hierarchical rubrics.
Qwen 3.8 Max 93%
- MLS-Bench In use Agent must invent one algorithmic change in an ML research scaffold whose gain transfers across settings. Scored relative to human baselines. Claude Fable 5.1 50.3%
- PostTrainBench 1.2 In use CLI agent gets a base model, one H100 and a time budget to post-train it for a target benchmark. Scored on normalized gain. Claude Fable 5.1 44.6%
- MLE-bench In use Kaggle-style ML engineering competitions. Agent trains models and is scored against competition leaderboards. GPT-6.2 Astra 96.1%
- PMPP-Hard In use Agents write GPU kernels for 69 tasks that must be correct, and on 48 of them also run within 1.25 times the reference time. The score is the area under the solved-tasks curve over a 10-hour budget, out of 690 task-hours. Kimi K3 0.71
- KernelBench In use Replace PyTorch operators with faster custom GPU kernels. Scored on correctness and speedup.
MiniMax-M3 28.8%
Data audit
All 10 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
- Pass Independent runs Scores from the official leaderboard (44 entries).
- Pass Models scored 26 models we track, plus 18 other entries on its leaderboard.
- Pass Newest score The newest is from November 2026, for Claude Fable 5.2.
- Pass Frontier models 5 of the top 10 have a score, led by Claude Fable 5.2 (#2).
- Note Headroom Scored as a speedup, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on inferencebench.ai: 53 rows for models we track at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.