The Trace Manifest
Subscribe
Corrections Subscribe

Kimi K3

Moonshot AI · Open-weight · K3

Widely considered the strongest open-weights model for general quality and complex agentic tasks.

Status
active
Release date
2026-07-01
Parameters
unknown
Context window
1M
Modalities
text,code
Capabilities recorded
tool use, API, local use
Last verified
2026-07-20

Reviewed use cases

  • complex agentic tasks
  • coding
  • reasoning

Recorded limitations

  • Smaller ecosystem than Llama/Qwen; limited third-party evaluation

Benchmark Performance

64.4 / 100   TRACE Aggregate Score

TRACE Score 64.4/100 based on 2 benchmarks. 1 independent run (weighted 3×). 1 vendor-reported run (weighted 1×). Last updated 01/07/2026.

BenchmarkScoreProvenanceDateSource
SWE-bench Verified ~70% Vendor 01/07/2026 Source
LMSYS Chatbot Arena ~1300 Independent 01/07/2026 Source

Pricing is omitted until provider-model pricing rows have their own completed review and publication state.