Know the exact model.
The canonical record is moonshotai:kimi-k2-thinking. Capabilities and context limits reflect source observations; the limits of a particular inference endpoint may differ.
Inclusion in the radar does not mean this model is served by Flonno. For the model currently available through our API, see Qwen3.8 Flash 27B and API pricing.
Artifact evidence
moonshotai/Kimi-K2-Thinking@a51ccc050d73dab088bf7b0e2dd9b30ae85a4e55
official model card · Pinned artifact · Hugging Face Hub. First observed Oct 12, 2026; last observed Oct 13, 2026.
moonshotai/Kimi-K2-Thinking
official model card · Mutable source entry · Hugging Face Hub. First observed Oct 12, 2026; last observed Oct 13, 2026.
artificial-analysis:bddebfd3-0a8d-47f5-b722-bc4c2ca5a5dc
independent benchmark · Pinned artifact · Artificial Analysis evaluated configuration. First observed Oct 13, 2026; last observed Oct 13, 2026.
https://www.kimi.com/blog/kimi-k2-thinking|kimi-k2-thinking|release
official changelog · Pinned artifact · Official provider announcement. First observed Oct 12, 2026; last observed Oct 12, 2026.
Evaluate it for your workflow.
Pin the artifact and inference configuration. Use representative inputs with a written answer rubric, then record accepted results, errors, complete request time and token usage. Keep output budgets and concurrency explicit.
Benchmark results apply to their recorded version and method. They do not transfer automatically to a different runtime or a quantized derivative.
Explore practical comparison guides →