Benchmarks › FrontierCode
FrontierCode
Make repository changes that pass tests and a maintainer rubric
57.1top score · Claude Opus 5.5
28models ranked
15open-weights models
Mean rubric score with blocker gatinghigher is better
Oct 20, 2026last updated
How the test works
1
Task
100 tasks
- What the model doesMake repository changes that pass tests and a maintainer rubric
- Kinds of tasksSoftware engineering
- In → outText, repository → Code patch
2
Run
3 runs per task
- Agent / harnessStirrup
- Time limit4 h per task
- Step limit250 steps
3
Score
Mean rubric score with blocker gating
- Graded byTests and GPT-6 Luna (medium) rubric judge; blockers and trajectory flags score zero
- Better isHigher
Sources: artificialanalysis.ai, cognition.com
FrontierCode 1.1 leaderboardtop 10 of 28
Full leaderboard with filters →| Model | FrontierCode 1.1 | Per task | Price | Speed | Supports | Context |
|---|---|---|---|---|---|---|
57.1medium | $1.34 | $4 / $20 | 81 | 1M | ||
49.1max | $0.72 | $2 / $10 | 52 | 1.05M | ||
49max | $7.63 | $10 / $50 | 69 | 1M | ||
47max | $3.26 | $10 / $50 | 69 | 1.05M | ||
46.1xhigh | $2.01 | $2 / $10 | 112 | 1M | ||
45.5xhigh | $0.12 | $0.10 / $0.50 | 196 | 1M | ||
45.4thinking | $0.13 | $0.43 / $0.87 | 54 | 1M | ||
43.7max | $0.26 | $0.30 / $1.20 | 167 | 1M | ||
43.6max | $2.00 | $3 / $15 | 112 | 1M | ||
43.5high | $1.99 | $2 / $10 | 84 | 1M |
FrontierCode questions
What is FrontierCode?›
Make repository changes that pass tests and a maintainer rubric. It has 100 tasks. Created by Cognition.
Which model is best on FrontierCode?›
Claude Opus 5.5 leads FrontierCode 1.1 with 57.1, ahead of GPT-6.1 Sol (49.1) and Claude Fable 5.1 (49).
What is the best open-weights model on FrontierCode?›
MiMo v2.6 Pro with 45.4, then DeepSeek V4.1 Flash 1017 (43.7).
How is FrontierCode scored?›
Mean rubric score with blocker gating; tests and gpt-6 luna (medium) rubric judge; blockers and trajectory flags score zero; 3 runs per task per task.