- Cleus
- AI benchmarks
- FrontierSWE
FrontierSWEexplained
Software projects that take an agent up to 20 hours. FrontierSWE gives a coding agent very long software tasks, such as rebuilding a well-known tool in another language, speeding up slow scientific code or training a small forecasting model, with up to 20 hours per task. Every model runs in the same simple agent setup at its highest reasoning setting, and partial progress earns partial credit.
- Highest published score
- GPT-6 Astra, 65.5%
- Leader among popular models
- GPT-6 Astra
- Models with a result
- 21
- Skill it tests
- Coding
What FrontierSWE really tests
- What it is
- FrontierSWE gives a coding agent very long software tasks, such as rebuilding a well-known tool in another language, speeding up slow scientific code or training a small forecasting model, with up to 20 hours per task. Every model runs in the same simple agent setup at its highest reasoning setting, and partial progress earns partial credit.
- A question like
- “Reimplement the core of a version-control system in a different programming language so that its test suite passes.” Our own example in the style of the test, not a real item.
- Why it matters
- Most coding tests take minutes. This one asks whether a model can stay on track through a long project, which is where AI coding is heading.
- Reading the score
- Average share of each task completed (0–100%).
- The catch
- There are only a few dozen tasks, so one task can move a score noticeably. The first and second versions use different tasks and scales, so don’t compare scores across them.
- Made by
- Proximal, 2026. Official page
Leaderboard
Each model’s best published result across its reasoning settings.
Popular models
The models we track across this site.
Top 20 of all 21 models tested
Includes older, preview and specialist models that aren’t on our main leaderboard.
Questions
What is FrontierSWE?
FrontierSWE gives a coding agent very long software tasks, such as rebuilding a well-known tool in another language, speeding up slow scientific code or training a small forecasting model, with up to 20 hours per task. Every model runs in the same simple agent setup at its highest reasoning setting, and partial progress earns partial credit.
What does a FrontierSWE score mean?
Average share of each task completed (0–100%).
Which AI model is best on FrontierSWE?
GPT-6 Astra from OpenAI has the highest published FrontierSWE result we have (65.5%).
Is FrontierSWE a good benchmark?
Most coding tests take minutes. This one asks whether a model can stay on track through a long project, which is where AI coding is heading. The catch: There are only a few dozen tasks, so one task can move a score noticeably. The first and second versions use different tasks and scales, so don’t compare scores across them.
Who made FrontierSWE?
Proximal, 2026.
Benchmark scores: Epoch AI, Capabilities & benchmarking, used under CC BY 4.0. For each model we show its best published result across reasoning settings. Prices and context sizes: OpenRouter. Data refreshed November 9, 2026. Hands-on tests (answers, websites, code, refusals, speed) are our own single runs. Explanations written by Cleus.