← All benchmarks

PostTrainBench (Kimi K3 pre-v1.1 H20 evaluation)

Indexes · 2026-07-16

Kimi K3 release's historical PostTrainBench aggregate, averaged over three runs using the official Harbor implementation on H20 rather than the official H100 hardware. These July 16, 2026 results precede the July 28 v1.1 contamination, external-API and model-identity checks; keep the modified hardware and original scoring protocol distinct from official v1.1 and v1.2 results.

Top models (higher is better)