FrontierCode
Real-repo coding tasks graded for mergeability: correctness, regressions, lint, tests, scope and code quality. Failing any blocker scores zero.
Tasks written by 20+ open-source maintainers, 40+ hours each. Only sample tasks are public.
- Level
- Expert Specialists with years of training, often over hours of work
- Scored
- Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
- Answers
- Code A program or patch run against tests.
Claude Opus 4.8 leads with 34.3% among models released by May 2026
Best score per model from the benchmark's own leaderboard, read on 2 Nov 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.
The newest model with a score here came out in May 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6 Astra
- Claude Opus 5.5
- GPT-6.1 Terra
- GPT-6.1 Sol
- and 14 more
- 1 Claude Opus 4.8 xhigh 34.3%
- 2 GPT-5.5 xhigh 25.5%
- 3 Claude Opus 4.7 xhigh 23%
- 4 GPT-5.4 Mini xhigh 17.8%
- 5
Gemini 3.1 Pro high 16.7% - 6 Kimi K2.6 none 16%
- 7 Claude Sonnet 4.6 high 15.1%
- 8 Kimi K2.5 none 6.86%
- 9
MiniMax-M2.7 none 5.98% - 10 SWE-1.6 none 5.50%
- 11 MiniMax-M2.5 none 5.34%
- 12 Gemini 3.1 Flash-Lite low 4.84%
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Opus 4.8 | 34.3% | xhigh, Claude Code | May 2026 | Official leaderboard |
| 2 | GPT-5.5 | 25.5% | xhigh, Codex | Apr 2026 | Official leaderboard |
| 3 | Claude Opus 4.7 | 23% | xhigh, Claude Code | Apr 2026 | Official leaderboard |
| 4 | GPT-5.4 Mini | 17.8% | xhigh, Codex | Mar 2026 | Official leaderboard |
| 5 | Gemini 3.1 Pro | 16.7% | high, Gemini CLI | Feb 2026 | Official leaderboard |
| 6 | Kimi K2.6 | 16% | none, mini-swe-agent | Apr 2026 | Official leaderboard |
| 7 | Claude Sonnet 4.6 | 15.1% | high, Claude Code | Feb 2026 | Official leaderboard |
| 8 | Kimi K2.5 | 6.86% | none, mini-swe-agent | Jan 2026 | Official leaderboard |
| 9 | MiniMax-M2.7 | 5.98% | none, mini-swe-agent | Mar 2026 | Official leaderboard |
| 10 | SWE-1.6 | 5.50% | none, Devin | Apr 2026 | Official leaderboard |
| 11 | MiniMax-M2.5 | 5.34% | none, mini-swe-agent | Feb 2026 | Official leaderboard |
| 12 | Gemini 3.1 Flash-Lite | 4.84% | low, Gemini CLI | Mar 2026 | Official leaderboard |
Epoch AI has scores for 46 models. Labs report scores for 4 models in their cards. We plot every source in the charts below.
The best score, 34.3%, costs $6.69 a task
Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| Claude Opus 4.8 | xhigh | Claude Code | 34.3% | $6.69 | 62,445 |
| Claude Opus 4.8 | max | Claude Code | 31.3% | $9.03 | 88,169 |
| Claude Opus 4.8 | high | Claude Code | 30.3% | $3.94 | 37,280 |
| Claude Opus 4.8 | medium | Claude Code | 26.9% | $3.30 | 31,701 |
| GPT-5.5 | xhigh | Codex | 25.5% | $5.59 | 25,576 |
| GPT-5.5 | high | Codex | 25.3% | $4.38 | 19,002 |
| Claude Opus 4.8 | low | Claude Code | 25.3% | $2.68 | 23,239 |
| Claude Opus 4.7 | xhigh | Claude Code | 23% | $5.93 | 35,077 |
| GPT-5.5 | medium | Codex | 22.4% | $3.29 | 13,912 |
| Claude Opus 4.7 | high | Claude Code | 22.1% | $4.43 | 28,180 |
| Claude Opus 4.7 | medium | Claude Code | 21.3% | $2.85 | 20,017 |
| GPT-5.5 | low | Codex | 21.1% | $1.61 | 8,482 |
| Claude Opus 4.7 | low | Claude Code | 17.9% | $2.43 | 16,932 |
| GPT-5.4 Mini | xhigh | Codex | 17.8% | $1.67 | 88,550 |
| Gemini 3.1 Pro | high | Gemini CLI | 16.7% | $2.41 | 18,997 |
| Gemini 3.1 Pro | low | Gemini CLI | 16.5% | $2.52 | 19,651 |
| Kimi K2.6 | none | mini-swe-agent | 16% | 41,815 | |
| Claude Sonnet 4.6 | high | Claude Code | 15.1% | $2.40 | 34,334 |
| Claude Sonnet 4.6 | medium | Claude Code | 13.6% | $2.09 | 28,176 |
| Claude Sonnet 4.6 | xhigh | Claude Code | 13.4% | $2.48 | 34,856 |
| Claude Sonnet 4.6 | max | Claude Code | 13.2% | $2.57 | 38,004 |
| GPT-5.4 Mini | high | Codex | 12.4% | $1.05 | 45,131 |
| GPT-5.4 Mini | medium | Codex | 12.1% | $0.89 | 34,720 |
| Kimi K2.5 | none | mini-swe-agent | 6.86% | $2.55 | 19,108 |
| MiniMax-M2.7 | none | mini-swe-agent | 5.98% | 30,587 | |
| SWE-1.6 | none | Devin | 5.50% | ||
| MiniMax-M2.5 | none | mini-swe-agent | 5.34% | 21,691 | |
| Gemini 3.1 Flash-Lite | low | Gemini CLI | 4.84% | $0.17 | 13,559 |
| Gemini 3.1 Flash-Lite | high | Gemini CLI | 4.44% | $0.18 | 14,679 |
| Gemini 3.1 Flash-Lite | minimal | Gemini CLI | 2.51% | $0.37 | 11,463 |
| Gemini 3.1 Flash-Lite | medium | Gemini CLI | 1.84% | $0.26 | 10,006 |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard, Epoch AI’s runs and what labs claim in their cards.
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
One lab put it in a launch post in the last six months
Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.
| Date | Launch post | Score | As the post writes it |
|---|---|---|---|
| 22 Sep 2026 | Claude Opus 5.5 | 54.4 | Agentic coding FrontierCode v1.1 (Main) |
| 24 Jul 2026 | Claude Opus 5 | 53.4 | Agentic coding FrontierCode v1.1, Main |
| 30 Jun 2026 | Claude Sonnet 5 | 38.8 | FrontierCode v1 |
| 9 Jun 2026 | Claude Fable 5 | 29.3 | Agentic coding FrontierCode (Diamond) (xhigh) |
2 system cards report it, from 1 lab
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| 30 Jun 2026 | Claude Sonnet 5 system card | 38.8 | FrontierCode v1, 150 tasks, max effort (xhigh 34.0%, high 28.9%) (Table 8.1.A, Figure 8.4.A) | Capabilities |
| 9 Jun 2026 | Claude Fable 5 system card | 29.3 | FrontierCode (Diamond subset), score at xhigh effort, mean@5, 150-task benchmark by Cognition; pass rate 30.2%; Main subset score 46.3% and pass rate 48.8% (Table 8.1.A, 8.4) | Capabilities |
Versions
We switch the top row to each new version when it comes out. Open it for one row per version. An old version stays in play while people still test it, so whoever first takes it past 85% gets credit for closing it out.
More in Coding
- SWE-bench Verified In use 500 human-validated GitHub issues from 12 Python repos. Agent writes a patch that must pass hidden tests. Claude Opus 4.7 83.5%
- SWE-bench Multilingual In use 300 GitHub-issue tasks across nine programming languages in SWE-bench format. Patch graded by repository tests. Claude Opus 5.5 93.9%
- LiveCodeBench v6 In use Competitive programming problems collected after model cutoffs from LeetCode, AtCoder, Codeforces. Scored pass@1 on hidden tests.
DeepSeek-V4-Pro 93.5% - DeepSWE In use 113 original long-horizon engineering tasks across 91 repos and five languages, written from scratch. Hand-written verifiers accept alternative solutions. GPT-6 Astra 74.1%
- SciCode In use Scientist-curated research coding problems split into subproblems across natural sciences. Code checked against test cases.
Seed2.1 Pro 59.8% - Codeforces rating In use Model solves recent Codeforces contest problems. An Elo-style rating is computed from simulated contest performance. DeepSeek V4.1 Flash 3,471