Board/Moonshot/Kimi K3
K

Kimi K3

Moonshot AI·Jul 22, 2026N/A

Open weights

Moonshot's open-frontier flagship: the first open model at 2.8T parameters, third place on major intelligence indexes (#5 of 226 tracked models), and the highest self-reported SWE-bench Verified figure we track (93.4%).

Status
Generally available
License
Open weights · Custom
Context window
1M
Max output
128K
Input · $/M
$3
Output · $/M
$15
Cached input · $/M
$0.3
Blended · $/M
$6.00
Parameters
2.8T MoE — first open model at this scale
Inputs
Text + images

Benchmarks

Composite 93.4· mean of published scores

Reasoning & knowledge

  • GPQA Diamond

    no public score

  • MMLU-Pro

    no public score

  • Humanity's Last Exam

    no public score

  • ARC-AGI-2

    no public score

Coding

  • SWE-bench VerifiedLeader

    93.4%

  • SWE-bench Pro

    no public score

  • LiveCodeBench

    no public score

Agents & computer use

  • Terminal-Bench 2

    no public score

  • OSWorld 2.0

    no public score

  • BrowseComp

    no public score

Math

  • AIME 2026

    no public score

Known strengths

  • Top self-reported SWE-V (93.4%)
  • 3rd on composite intelligence indexes
  • Flat pricing across the full 1M context

Caveats

  • 93.4% SWE-V is vendor-reported; independent reproduction is still limited
  • Hosted prices vary widely by provider ($3–$5 input)
  • Weights rollout staggered after launch

Primary sources

  • Kimi K3 tech blog
Compare this model →

More from Moonshot

  • KKimi K2.6Apr 8, 2026 · 256K ctx
  • KKimi K2 ThinkingNov 6, 2025 · 256K ctx