Kimi Hallu Bench: Results as Published

Kimi Hallu Bench is an in-house factual-accuracy test of Moonshot AI. The prompts are challenging, de-identified requests drawn from practical work tasks, and fact-checking agents check every factual claim in the answer. The score is the claim-level hallucination rate, the share of claims that contain an error, so lower is better. A Hard split with tools is built from examples crafted to probe situations where models are prone to errors.

Every number here is read from the same catalog file as the benchmark table, data version 2026103001, and is the number the table shows for that model and test. Where one page printed a model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.