Enterprise AI in Numbers · Capability and cost
Top score on FrontierSWE, the benchmark of long software engineering projects
65.5%
65.5% (GPT-6 Astra) as of September 3, 2026, according to Epoch AI, Benchmarking hub, up from 56.3% on September 1, 2026. The highest share of FrontierSWE tasks (long software engineering projects judged on implementation, performance and research) solved by any evaluated model, with the model and lab that hold it.
What it means for leaders
Coding is where model choice pays back fastest, and the lead changes hands within weeks. A new leader is a reason to rerun your own evaluation on your codebase, not to switch vendors on the spot.
What would change the reading: A lead held by a lab outside the top three, or an open-weight model within a few points of the top.
The trend
Every reading
Definition and method
- What it measures
- The highest share of FrontierSWE tasks (long software engineering projects judged on implementation, performance and research) solved by any evaluated model, with the model and lab that hold it.
- Population
- AI models evaluated on FrontierSWE, as compiled by Epoch AI
- Updated
- When new models are evaluated
- Source
- Epoch AI, Benchmarking hub · CC BY 4.0
Caveats
- Scores come from independent evaluators compiled by Epoch AI, not from the labs, and depend on the settings and tools each run used.
- The leader is the best result so far, dated by the release of the model that set it; a model evaluated late can take the lead after its release date.
- Each model is scored with the best agent harness it was run in.
Cite this
Enterprise AI in Numbers, "Top score on FrontierSWE, the benchmark of long software engineering projects", https://lucianoayres.com/ai-brief/data/coding-benchmark-leader/, retrieved October 25, 2026. Data: Epoch AI, Benchmarking hub.