OpenRouter Benchmark Harness

OpenRouter's internal benchmarking harness, externalized for transparency. We port benchmarks here so we can run them scalably on our infrastructure and iterate quickly.

bun install
OPENROUTER_API_KEY=... bun run bench -- --benchmark gpqa_diamond --model openai/gpt-4o-mini --limit 5

See CONTRIBUTING.md before proposing changes. Report security issues privately as described in SECURITY.md.