FrontierCode leaderboard
As of November 8, 2026, Claude Opus 5.5 (Anthropic) leads FrontierCode at 54.4%, ahead of GPT-6 Astra at 53.3% and Claude Sonnet 5.5 at 52.1%. 15 systems ranked. Every score here is published by the model's own lab; each row links to the page it comes from.
What is FrontierCode? FrontierCode judges coding agents the way an open-source maintainer would: on whether the change they submit would be merged, not only whether tests pass. Made by Cognition, with tasks built by 20+ open-source maintainers, June 2026 (v1.1 in July 2026).
Top models on FrontierCode, November 2026
| # | Model | Score | $ / task | Price / 1M | Context | Source |
|---|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 Anthropic | adaptive thinking, max effort | — | $4 / $20 | 1M | Vendor-reported Anthropic |
| 2 | GPT-6 Astra OpenAI | max effort | $4.59 | $10 / $50 | 1.05M | Vendor-reported OpenAI |
| 3 | Claude Sonnet 5.5 Anthropic | xhigh effort | — | $2 / $10 | 1M | Vendor-reported Anthropic |
| 4 | Claude Fable 5.1 Anthropic | as reported | — | $10 / $50 | 1M | Vendor-reported Anthropic |
| 5 | GPT-6 Sol OpenAI | max effort | $2.14 | $2 / $10 | 1.05M | Vendor-reported OpenAI |
| 6 | Claude Opus 5 Anthropic | as reported | — | $5 / $25 | 1M | Vendor-reported Anthropic |
| 7 | GPT-5.6 Sol OpenAI | max effort | $5.19 | $4 / $20 | 1.05M | Vendor-reported OpenAI |
| 8 | Claude Haiku 5.5 Anthropic | best reported effort | — | $0.1 / $0.5 | 1M | Vendor-reported Anthropic |
| 9 | Gemini 3.7 Flash Deprecated Google DeepMind | score | — | $0.75 / $3.75 | 1M | Vendor-reported Google DeepMind |
| 10 | Claude Sonnet 5 Anthropic | best reported effort | — | $2 / $10 | 1M | Vendor-reported Anthropic |
| 10 | GPT-6 Luna OpenAI | max effort | $0.11 | $0.1 / $0.5 | 1.05M | Vendor-reported OpenAI |
| 12 | GPT-5.6 Luna OpenAI | max effort | $0.37 | $0.2 / $1.2 | 1.05M | Vendor-reported OpenAI |
| 13 | Gemini 3.6 Flash Google | score | — | $0.75 / $3.75 | 1M | Vendor-reported Google DeepMind |
FrontierCode under other settings
Labs report FrontierCode under different settings, and a score from one setting is not comparable with a score from another. Each setting has its own table, ranked on its own.
v1.1 Extended (all 150 tasks)
| # | Model | Score | Price / 1M | Context | Source |
|---|---|---|---|---|---|
| 1 | GPT-6 Astra OpenAI | as reported | $10 / $50 | 1.05M | Vendor-reported OpenAI |
| 2 | Grok 4.6 SpaceXSI | high effort | $2 / $6 | 500K | Vendor-reported SpaceXSI (xAI) |
| 3 | GPT-5.6 Sol OpenAI | as reported | $4 / $20 | 1.05M | Vendor-reported OpenAI |
| 4 | Grok 4.5 SpaceXSI | high effort | $2 / $6 | 500K | Vendor-reported SpaceXSI (xAI) |
FrontierCode: score against price
What more reasoning buys
Some publishers ran the same model at several reasoning-effort levels and printed the score and the cost per task for each. Higher effort spends more tokens per task; this is what it bought on FrontierCode.
| Model | Score by effort level ($ per task) | Source |
|---|---|---|
| GPT-6 Astra | low 45.3% $1.70 medium 48.8% $2.43 high 50.9% $3.01 xhigh 50.6% $3.28 max 53.3% $4.59 | OpenAI |
| GPT-6 Sol | low 37.3% $0.45 medium 45.9% $0.80 high 47.7% $1.08 xhigh 48.4% $1.37 max 49.3% $2.14 | OpenAI |
| GPT-5.6 Sol | low 35.4% $1.89 medium 39.9% $2.69 high 45.1% $3.48 xhigh 46.8% $4.15 max 47.5% $5.19 | OpenAI |
| GPT-6 Luna | low 25.7% $0.021 medium 35.5% $0.053 high 37.3% $0.067 xhigh 37.1% $0.073 max 42.4% $0.11 | OpenAI |
| GPT-5.6 Luna | low 15.4% $0.060 medium 25.7% $0.13 high 35.9% $0.23 xhigh 38.9% $0.31 max 39.8% $0.37 | OpenAI |
Can you trust FrontierCode scores?
No independent leaderboard publishes FrontierCode results for these models yet, so every number on this page is the lab's own report of its own model. Labs choose the harness, the effort level and the subset they report, and they rarely publish failures. Read the setting under each score before comparing two rows.
- The tasks are private to avoid contamination, so no one outside Cognition can reproduce a score. Cognition also builds a coding agent of its own.
- Agents have internet access and could find the upstream fix; v1.1 zeroes runs a scanner flags instead of blocking the network.
- Rubric grading involves judgement; v1.1 relaxed 75 overly strict blocker criteria, so v1.0 and v1.1 scores differ.
How FrontierCode works
The task
Each task is a realistic change request in a maintainer's own repository. The agent submits a solution that is reviewed for correctness, tests, scope and style.
The score
A weighted score over reviewer-defined rubric items, unit tests and verifiers, where failing any blocker criterion scores zero. Each model runs five times per effort level and the average at its best effort is reported.
Test set
150 tasks; Main is the 100 hardest, as published by the authors.
Made by
Cognition, with tasks built by 20+ open-source maintainers, first published June 2026 (v1.1 in July 2026).
Official sources
- cognition.com/blog/frontier-code-1.1
- Definition read on 2026-11-08
Sources and freshness
17 published scores from 7 sources. Each was read on its source page; the oldest reading on this page is from November 8, 2026. Prices beside each score come from our model price index and move when a vendor reprices.
- Vendor-reported OpenAI
- Vendor-reported Anthropic
- Vendor-reported Anthropic
- Vendor-reported Google DeepMind
- Vendor-reported OpenAI
- Vendor-reported SpaceXSI (xAI)
- Vendor-reported Anthropic
FrontierCode FAQ
Answers derived from the 17 published scores on this page and the current price index.