FrontierCode
Frontier-difficulty programming problems.
Agent Index weight
0%
shown for context only
Models tested
46
from the benchmark’s leaderboard and reports
Results
| # | Model | Score | Normalised | Harness | Cost/task | Date |
|---|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 Supported Anthropic | 54.6% | claude-code | — | — | |
| 2 | Claude Fable 5 Supported Anthropic | 53.5% | claude-code | — | — | |
| 3 | Claude Opus 5 Supported Anthropic | 53.4% | claude-code | — | — | |
| 4 | GPT-6 Astra Supported OpenAI | 53.3% | codex | — | — | |
| 5 | Claude Sonnet 5.5 Estimated Anthropic | 52.1% | claude-code | — | — | |
| 6 | Claude Fable 5.1 Supported Anthropic | 50.9% | claude-code | — | — | |
| 7 | GPT-6.1 Sol Estimated OpenAI | 50.2% | codex | — | — | |
| 8 | SWE-2 Cognition | 50.0% | devin | — | — | |
| 9 | GPT-6 Sol Estimated OpenAI | 49.3% | codex | — | — | |
| 10 | Grok 4.6 Supported xAI | 48.0% | grok-build | — | — | |
| 11 | Grok 4.7 Estimated xAI | 47.6% | grok-build | — | — | |
| 12 | Gemini 4 Argon Supported Google | 47.5% | devin-cli | — | — | |
| 13 | GPT-5.6 Sol Supported OpenAI | 47.5% | codex | — | — | |
| 14 | Claude Opus 4.8 Supported Anthropic | 46.5% | claude-code | — | — | |
| 15 | Claude Haiku 5.5 Estimated Anthropic | 46.4% | claude-code | — | — | |
| 16 | Kimi K3 Supported Moonshot AI | 44.2% | mini-swe-agent | — | — | |
| 17 | DeepSeek V4.1 Flash Estimated DeepSeek | 43.9% | devin-cli | — | — | |
| 18 | Gemini 3.7 Flash Estimated Google | 43.6% | devin-cli | — | — | |
| 19 | GPT-5.5 Supported OpenAI | 43.0% | codex | — | — | |
| 20 | Claude Sonnet 5 Supported Anthropic | 42.7% | claude-code | — | — | |
| 21 | Grok 4.5 Estimated xAI | 42.4% | grok-build | — | — | |
| 22 | GPT-6 Luna OpenAI | 42.4% | codex | — | — | |
| 23 | SWE-1.7 Cognition | 42.0% | devin-cli | — | — | |
| 24 | GPT-5.6 Terra Estimated OpenAI | 41.3% | codex | — | — | |
| 25 | Gemini 3.8 Flash Supported Google | 41.2% | devin-cli | — | — | |
| 26 | GLM-5.3 Supported Z.ai (Zhipu) | 40.1% | devin-cli | — | — | |
| 27 | GPT-5.6 Luna Estimated OpenAI | 39.8% | codex | — | — | |
| 28 | Mistral Large 4 Mistral | 39.4% | chisel | — | — | |
| 29 | Claude Opus 4.7 Supported Anthropic | 38.5% | claude-code | — | — | |
| 30 | Gemini 3.6 Flash Estimated Google | 34.4% | devin-cli | — | — | |
| 31 | GLM-5.3-Flash Supported Z.ai (Zhipu) | 31.8% | devin-cli | — | — | |
| 32 | Kimi K2.7 Code Estimated Moonshot AI | 30.1% | mini-swe-agent | — | — | |
| 33 | DeepSeek V4 Pro 0813 DeepSeek | 28.5% | devin-cli | — | — | |
| 34 | GPT-5.4 mini OpenAI | 27.0% | codex | — | — | |
| 35 | Claude Opus 4.6 Supported Anthropic | 26.6% | claude-code | — | — | |
| 36 | Composer 2.5 Cursor | 25.6% | cursor-cli | — | — | |
| 37 | GLM-5.2 Supported Z.ai (Zhipu) | 24.5% | mini-swe-agent | — | — | |
| 38 | Claude Sonnet 4.6 Supported Anthropic | 24.3% | claude-code | — | — | |
| 39 | DeepSeek V4 Flash 0731 DeepSeek | 18.8% | devin-cli | — | — | |
| 40 | DeepSeek V4 Pro Estimated DeepSeek | 17.6% | mini-swe-agent | — | — | |
| 41 | MiniMax-M3 Estimated MiniMax | 14.7% | mini-swe-agent | — | — | |
| 42 | Inkling Estimated thinkingmachines | 14.0% | mini-swe-agent | — | — | |
| 43 | Nemotron 3 Ultra NVIDIA | 13.6% | devin-cli | — | — | |
| 44 | Qwen3.7 Plus Alibaba (Qwen) | 10.2% | mini-swe-agent | — | — | |
| 45 | SWE-1.6 Cognition | 9.4% | devin-cli | — | — | |
| 46 | Mistral Medium 3.5 Mistral | 8.0% | devin-cli | — | — |