benchmark

Coming soon

We evaluate frontier models on challenging real-world tasks: same prompts, same harness, no vendor tuning, reproducible.