Benchmarks

Models, raced head-to-head

When a new model ships, we race it against the field on real agentic tasks — same challenge, same tools — and score the whole trajectory on correctness, reliability, latency, and cost. Here is who won.