SYLIN

Kimi

by Moonshot AI · Model · Language & code models

strong

The leading open-weights model for agents and web development, but enormous to run.

#1 agents (open weights)#1 documents (open weights)#1 web development (open weights)

#1 on Arena agent among open weights

AccessOpen weightsNeeds~1.56 TB; smallest build ~610 GBDownloadabout 1.56 TB full weights; one GGUF needs about 610 GBSize2.78T parametersLicenseKimi K3.1 Pro License

Why we say so

Arena ranks K3.1 first among open weights on agents and web development, and Artificial Analysis places it above Fable 5.1 and GPT-6 at about $2 per task. It also beat Qwen 4 Max on Terminal-Bench (67% versus 48%). The claim of near-frontier performance is Moonshot's own, and the weights are 1.56 TB, so most people will use an API or a heavily compressed quantization.

Versions

VersionReleasedRank
Kimi K3.1LATEST
strong Top open-weights model on Arena agents and web development
Oct 25#1 agents (open weights)#1 web development (open weights)
Kimi K3Jul 16#3 chat (open weights)
Kimi K2.6
promising First among open weights on Arena documents, second on vision
#1 documents (open weights)#2 image understanding (open weights)
Kimi K2.5

Evidence

  1. Nov 18
    vLLM reports up to 30% higher acceptance length from a sliding-window draft for Kimi K3.1 Pro
  2. Nov 15
    #1 on Arena webdev among open weights (1684)
  3. Nov 11
    #3 on Arena text among open weights (1504)
  4. Nov 11
    #1 on Arena webdev among open weights (1687)
  5. Nov 9
    #1 on Arena document among open weights (1450)
Show all 39
  1. Nov 9
    #2 on Arena vision among open weights (1282)
  2. Nov 8
    Best-performing open model across agentic coding, general knowledge and science benchmarks
  3. Nov 8
    Full open weights released: 2.8T total, 104B active MoE, 1.56 TB; described as best open model, near top closed models
  4. Nov 6
    Took #1 on Arena webdev among open weights
  5. Oct 25
    Kimi K3.1 Pro released: 3T parameters, 1M context, on par with frontier models; open weights promised November 4
  6. Oct 25
    Ranks above Claude Fable 5.1, GPT-6 and Gemini 4 on Artificial Analysis index at about $2 per task
  7. Oct 25
    Highest Terminal-Bench score at lower cost per task than GPT and Claude, per vendor chart
  8. Oct 19
    #2 on Arena vision among open weights (1282)
  9. Oct 16
    Took #1 on Arena webdev among open weights
  10. Sep 27
    2.78T-parameter MoE runs at about 1 tok/s on an M5 Max MacBook Pro (128 GB) with experts streamed from four SSDs
  11. Sep 13
    #1 on Arena document among open weights (1451)
  12. Sep 9
    Reported in the stolen-thoughts paper to get much closer to Claude 4.8 answers when prefilled with its reasoning, hinting at distillation
  13. Aug 28
    Found unpatched Mali GPU CVE-2022-38181 on the tablet's firmware, the starting point for the root exploit
  14. Aug 22
    Kimi K3 open weights reached 2,930 (52.2% gap closed) with Prime Agent harness; 2,974 with kimi-code, near Opus 5
  15. Aug 9
    Trained with full thinking history preserved; harnesses like Hermes that resend reasoning can destabilise quality, so Moonshot recommends its own harness
  16. Aug 4
    Fixed all bugs in about 8-9 minutes with ~17 live assertions, the most thorough checks, but left a stray booking uncleaned
  17. Aug 2
    Full weights released on Hugging Face on July 27 as promised: 2.8T MoE, 104B active, native vision (MoonViT)
  18. Jul 25
    Kebab prompt took 1h16m, about three times Opus 5's time; cheaper per token but similar total cost expected; thinner execution
  19. Jul 20
    Map-image geography and multilingual task: correct Netherlands ID, Dutch/Arabic/Portuguese lines, about 40 seconds with short reasoning
  20. Jul 19
    Output price a bit over three times cheaper than Fable 5; per launch posts K3 leads on Terminal-Bench, AutomationBench, GPQA, while Fable leads others
  21. Jul 19
    Beats Opus 4.8 on KernelBench CUDA; on KernelBench hard it is below Fable 5 but above Opus 4.8
  22. Jul 19
    Autonomously designed and optimized a small inference chip for a smaller model over 48 hours, in simulation
  23. Jul 17
    Moonshot launched Kimi K3, a 2.8T-parameter model with 1M-token context and native vision; API live, weights not yet public
  24. Jul 17
    Moonshot's own blog admits a noticeable gap to Claude and GPT frontier models; claims it beats Fable and Sol were refuted
  25. Jul 17
    Leads open models on Artificial Analysis index, about two points behind the best GPT, ahead of GLM 5.2
  26. Jul 17
    DeepSWE close to Fable 5 and GPT 5.6, over 20 points above Opus 4.8 and the next best open model
  27. Jul 17
    Cost-versus-performance charts place K3 as more cost-efficient per task than Claude Mythos and Opus
  28. Jul 17
    Ranked #1 on front-end code arena with 1679 points, ahead of Fable 5; #1 in six of seven front-end domains, #2 in gaming
  29. Jul 17
    Liquid splash simulator with webcam hand tracking built from scratch in one prompt; ran 30 minutes, 2M input tokens, 46K output
  30. Jul 17
    On the same prompt, Kimi K3 gave the richest simulation, live network animation, grounded assertions; reviewer rated it best, above Fable 5
  31. Jul 16
    Reported top on ProgramBench and SWE Marathon, beating Opus, both GPT models and GLM
  32. Jul 16
    Roughly tied with GPT Sol on Terminal-Bench, ahead of Opus and Fable
  33. Apr 22
    Took #1 on Arena document among open weights
  34. Jan 29
    Took #1 on Arena vision among open weights

Cost to run

Alternatives

GLM#2 on Arena text among open weights (1509)
Claude FableIdentified 2 of 6 brain tumor types correctly, versus 1 of 6 for GPT-6 Astra and Kimi; it answered despite earlier Fable refusing medical questions
GPT SolEven after the cut, Grok 4.6 at $6/M offers roughly similar intelligence cheaper; Sol is now cheaper than Kimi K3
Claude OpusTop of Terminal-Bench-Science at 30.0% resolution with Claude Code, ahead of GPT Sol 22.4% and Fable 21.4%
DeepSeek Flash#2 on Arena webdev among open weights (1659)
Qwen MaxScores 48% on Terminal-Bench 4.0 against Kimi K3.1 Pro's 67%; Kimi costs $3.68 per index task
Made for Kimi