Open benchmarks
How Kodwen compares on the benchmarks the market actually quotes. Our own numbers come from runs through the gateway — the same path your requests hit. Each block is labelled: official harness, proxy, internal suite, or published reference scores we did not run ourselves.
Loading benchmark results…
No public benchmarks yet
Runs are published as they complete. Check back shortly.
Methodology
ai.codai.ro with a real API key. There is no internal fast path: measurements include the same auth → router → cascade → upstream → usage rollup path your requests hit.usage_events table billing pulls from. Cost per successful task is the only metric that survives gaming — it punishes both expensive correct answers and cheap wrong answers.