agents benchmark
Frontier-Bench v0.1
Frontier-Bench v0.1 evaluates agentic terminal coding. Anthropic reports results using the mini-SWE-agent harness and a GKE backend, measured as mean reward across five attempts per task.
Updated Aug 11, 2026
Models1
Model coverage1
MetricScore
EvidenceB
Frontier-Bench v0.1 Ranking
Higher score ranks better on this benchmark.
| 1 | AN | 43.3% | 100.0% | 1 | C |
Frontier-Bench v0.1 Highlights
The leading models and scores on this benchmark.
What is Frontier-Bench v0.1?
What Frontier-Bench v0.1 measures and how its scores work.
Frontier-Bench v0.1 evaluates agentic terminal coding. Anthropic reports results using the mini-SWE-agent harness and a GKE backend, measured as mean reward across five attempts per task.
Scores are shown in ratio. This benchmark is not independently verified and has an evidence level of B.
- Family
- Frontier-Bench v0.1
- Modality
- text
- Primary category
- agents
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- frontier-bench-v0.1|llm-stats-current
Benchmark scores retain their original unit. Overall score eligibility is shown separately.
FAQ
Common questions about Frontier-Bench v0.1.
Which model scores highest on Frontier-Bench v0.1?
Claude Opus 5 is currently ranked first with 43.3%.
What does Frontier-Bench v0.1 measure?
Frontier-Bench v0.1 evaluates agentic terminal coding. Anthropic reports results using the mini-SWE-agent harness and a GKE backend, measured as mean reward across five attempts per task.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
1 model results are currently shown.
Does this benchmark affect the overall score?
No. This benchmark is shown for reference but does not contribute to the overall score.