Benchmarks for frontier models
Hard, reproducible evaluations that show where today’s best models fall short, and why.
Evaluations for frontier models
From public benchmarks to private evaluations built around your roadmap.
- 01
Frontier benchmarks
Hard tasks that separate today’s best models.
- 02
Custom evaluations
Built around the skills you’re training next.
- 03
Agent evaluations
Long, tool-using tasks scored end to end.
Frontier models on expert work
Top models now score 97% on older coding benchmarks. On expert work in finance, science and medicine, the best model still fails a third to half of the tasks.
- SWE-bench VerifiedFor contrast: an older benchmark
- Vals Science IndexLife sciencesTop models1. GPT-6.2 Astra68.4%2. GPT-6 Astra65.1%3. Claude Fable 5.263.7%
- Terminal-Bench 4.0Software engineeringTop models1. Claude Fable 5.265.2%2. Claude Opus 5.565.2%3. Claude Sonnet 5.564.1%
- MedCodeMedicineTop models1. Claude Opus 563.6%2. Gemini 3.1 Pro Preview59.1%3. Gemini 4 Argon58.8%
- Tax Agent BenchFinanceTop models1. Gemini 4 Hydrogen52.3%2. Gemini 4 Carbon51.8%3. Claude Fable 5.150.8%
Public leaderboards from Vals AI, read November 17, 2026: SWE-bench Verified, Terminal-Bench 4.0, Tax Agent Bench, Vals Science Index, MedCode. Scores belong to their publishers. PrimeSet benchmarks are in progress.
Reproducible by design
Every score ships with the setup to reproduce it.
- 01Versioned tasksEvery run is pinned and repeatable.
- 02Graded in codeAnswers and evidence, never a judge’s opinion alone.
- 03Failure breakdownsSee where and why each model fails.
- 04Held-out testsKept out of training data.
Domains we cover
Built with the people who do the work in each field.
- Software engineeringRepositories, tests and tooling
- MathematicsProofs and quantitative reasoning
- Life sciencesBiology and chemistry
- Physical sciencesPhysics and engineering
- FinanceReports, models and filings
- LawContracts and case work
- MedicineRecords and guidelines
- Computer useBrowsers, apps and documents