FrontierCode
ReportedNo description yet: we only describe an eval set when we can do it correctly.
- Kind
- Published by a third party's leaderboard, collected by Epoch AI. The leaderboard is named and linked.
- Link
- https://cognition.com/frontiercode
- Score shown
- Epoch's column "Main score", as a percentage, higher is better; random guessing scores about 0.0%. The source gives no error bars.
- Results
- 46 in all: 14 for open-weight models, 31 for closed ones, 1 where Epoch does not say. 11 linked to a model page here.
Best open-weight result: moonshotai/Kimi-K3, 44.2%, number 16 of 46 results.
| # | Model | Weights | Setting | Score | Date | Source |
|---|---|---|---|---|---|---|
| 1 | claude-opus-5-5_mediumAnthropic | closed | effort medium; Harness: claude-code; Reasoning effort: medium | 54.6% | not dated | cognition.com |
| 2 | claude-fable-5_unknownAnthropic | closed | effort unknown; Harness: claude-code; Reasoning effort: xhigh | 53.5% | not dated | cognition.com |
| 3 | claude-opus-5_maxAnthropic | closed | effort max; Harness: claude-code; Reasoning effort: medium | 53.4% | not dated | cognition.com |
| 4 | gpt-6-astra_maxOpenAI | closed | effort max; Harness: codex; Reasoning effort: max | 53.3% | not dated | cognition.com |
| 5 | claude-sonnet-5-5_xhighAnthropic | closed | effort xhigh; Harness: claude-code; Reasoning effort: xhigh | 52.1% | not dated | cognition.com |
| 6 | claude-fable-5-1_mediumAnthropic | closed | effort medium; Harness: claude-code; Reasoning effort: medium | 50.9% | not dated | cognition.com |
| 7 | gpt-6.1-sol_mediumOpenAI | closed | effort medium; Harness: codex; Reasoning effort: medium | 50.2% | not dated | cognition.com |
| 8 | swe-2_maxCognition | closed | effort max; Harness: devin; Reasoning effort: max | 50.0% | not dated | cognition.com |
| 9 | gpt-6-sol_maxOpenAI | closed | effort max; Harness: codex; Reasoning effort: max | 49.3% | not dated | cognition.com |
| 10 | grok-4.6_unknownxAI | closed | effort unknown; Harness: grok-build; Reasoning effort: high | 48.0% | not dated | cognition.com |
| 11 | grok-4.7_unknownxAI | closed | effort unknown; Harness: grok-build; Reasoning effort: high | 47.6% | not dated | cognition.com |
| 12 | gemini-4-argon_mediumGoogle DeepMind | closed | effort medium; Harness: devin-cli; Reasoning effort: medium | 47.5% | not dated | cognition.com |
| 13 | gpt-5.6-sol_unknownOpenAI | closed | effort unknown; Harness: codex; Reasoning effort: max | 47.5% | not dated | cognition.com |
| 14 | claude-opus-4-8_unknownAnthropic | closed | effort unknown; Harness: claude-code; Reasoning effort: max | 46.5% | not dated | cognition.com |
| 15 | claude-haiku-5-5_maxAnthropic | closed | effort max; Harness: claude-code; Reasoning effort: max | 46.4% | not dated | cognition.com |
| 16 | moonshotai/Kimi-K3Epoch: kimi-k3_unknown, Moonshot | open | effort unknown; Harness: mini-swe-agent; Reasoning effort: none | 44.2% | not dated | cognition.com |
| 17 | deepseek-ai/DeepSeek-V4.1-FlashEpoch: deepseek-v4.1-flash_high, DeepSeek | open | effort high; Harness: devin-cli; Reasoning effort: high | 43.9% | not dated | cognition.com |
| 18 | gemini-3.7-flash_unknownGoogle DeepMind | closed | effort unknown; Harness: devin-cli; Reasoning effort: medium | 43.6% | not dated | cognition.com |
| 19 | gpt-5.5_unknownOpenAI | closed | effort unknown; Harness: codex; Reasoning effort: xhigh | 43.0% | not dated | cognition.com |
| 20 | claude-sonnet-5_unknownAnthropic | closed | effort unknown; Harness: claude-code; Reasoning effort: xhigh | 42.7% | not dated | cognition.com |
| 21 | grok-4.5_unknownxAI | closed | effort unknown; Harness: grok-build; Reasoning effort: high | 42.4% | not dated | cognition.com |
| 22 | gpt-6-luna_maxOpenAI | closed | effort max; Harness: codex; Reasoning effort: max | 42.4% | not dated | cognition.com |
| 23 | swe-1.7Cognition | not stated | Harness: devin-cli; Reasoning effort: none | 42.0% | not dated | cognition.com |
| 24 | gpt-5.6-terra_unknownOpenAI | closed | effort unknown; Harness: codex; Reasoning effort: max | 41.3% | not dated | cognition.com |
| 25 | gemini-3.8-flash_mediumGoogle DeepMind | closed | effort medium; Harness: devin-cli; Reasoning effort: medium | 41.2% | not dated | cognition.com |
| 26 | zai-org/GLM-5.3Epoch: glm-5.3_max, Z.ai (Zhipu AI) | open | effort max; Harness: devin-cli; Reasoning effort: max | 40.1% | not dated | cognition.com |
| 27 | gpt-5.6-luna_unknownOpenAI | closed | effort unknown; Harness: codex; Reasoning effort: max | 39.8% | not dated | cognition.com |
| 28 | mistral-large-4-0_highMistral AIno page match | open | effort high; Harness: chisel; Reasoning effort: high | 39.4% | not dated | cognition.com |
| 29 | claude-opus-4-7_unknownAnthropic | closed | effort unknown; Harness: claude-code; Reasoning effort: max | 38.5% | not dated | cognition.com |
| 30 | gemini-3.6-flash_unknownGoogle DeepMind | closed | effort unknown; Harness: devin-cli; Reasoning effort: medium | 34.4% | not dated | cognition.com |
| 31 | zai-org/GLM-5.3-FlashEpoch: glm-5.3-flash_max, Z.ai (Zhipu AI) | open | effort max; Harness: devin-cli; Reasoning effort: max | 31.8% | not dated | cognition.com |
| 32 | moonshotai/Kimi-K2.7-CodeEpoch: kimi-k2.7-code, Moonshot | open | Harness: mini-swe-agent; Reasoning effort: none | 30.1% | not dated | cognition.com |
| 33 | deepseek-ai/DeepSeek-V4-Pro-0813Epoch: deepseek-v4-pro-0813_high, DeepSeek | open | effort high; Harness: devin-cli; Reasoning effort: high | 28.5% | not dated | cognition.com |
| 34 | gpt-5.4-mini-2026-03-17_unknownOpenAI | closed | effort unknown; Harness: codex; Reasoning effort: xhigh | 27.0% | not dated | cognition.com |
| 35 | claude-opus-4-6_unknownAnthropic | closed | effort unknown; Harness: claude-code; Reasoning effort: high | 26.6% | not dated | cognition.com |
| 36 | Composer 2.5Cursor | closed | Harness: cursor-cli; Reasoning effort: none | 25.6% | not dated | cognition.com |
| 37 | zai-org/GLM-5.2Epoch: glm-5.2_none, Z.ai (Zhipu AI) | open | effort none; Harness: mini-swe-agent; Reasoning effort: none | 24.5% | not dated | cognition.com |
| 38 | claude-sonnet-4-6_unknownAnthropic | closed | effort unknown; Harness: claude-code; Reasoning effort: max | 24.3% | not dated | cognition.com |
| 39 | deepseek-ai/DeepSeek-V4-Flash-0731Epoch: deepseek-v4-flash-0731_unknown, DeepSeek | open | effort unknown; Harness: devin-cli; Reasoning effort: high | 18.8% | not dated | cognition.com |
| 40 | deepseek-ai/DeepSeek-V4-ProEpoch: deepseek-v4-pro_none, DeepSeek | open | effort none; Harness: mini-swe-agent; Reasoning effort: none | 17.6% | not dated | cognition.com |
| 41 | MiniMaxAI/MiniMax-M3Epoch: MiniMax-M3, MiniMax | open | Harness: mini-swe-agent; Reasoning effort: none | 14.7% | not dated | cognition.com |
| 42 | thinkingmachines/InklingEpoch: Inkling, Thinking Machines | open | Harness: mini-swe-agent; Reasoning effort: 0.99 | 14.0% | not dated | cognition.com |
| 43 | nemotron-3-ultraNvidiano page match | open | Harness: devin-cli; Reasoning effort: none | 13.6% | not dated | cognition.com |
| 44 | qwen3.7-plusAlibaba | closed | Harness: mini-swe-agent; Reasoning effort: none | 10.2% | not dated | cognition.com |
| 45 | swe-1.6Cognition | closed | Harness: devin-cli; Reasoning effort: none | 9.4% | not dated | cognition.com |
| 46 | mistral-medium-2604Mistral AIno page match | open | Harness: devin-cli; Reasoning effort: none | 8.0% | not dated | cognition.com |
Open-weight rows are shaded. Dates are when the result was run or evaluated, where the data says. Data from Epoch AI (CC-BY 4.0), fetched 19 Nov 2026: Epoch AI, ‘Capabilities & benchmarking’. Published online at epoch.ai. Retrieved from ‘https://epoch.ai/benchmarks’ [online resource]. Epoch AI. Method