Kimi K3 at a glance
Kimi K3 is a 2.8-trillion-parameter mixture-of-experts model with native vision and a 1M-token context, with weights released under the Kimi K3 License.
Current standing
As of
- Rank
- #38of 43 models
Kimi K3 (Low) · 17.4 on the Artificial Analysis Intelligence Index v5.0.1 (opens in a new tab)
- Current leader
- Gemini 4 Carbon (Google) · 60
Computed from our daily snapshot; it changes as new models are measured. Moonshot AI launches Kimi K3, a 2.8T-parameter open-weight model
What changed with Kimi K3
Moonshot AI published Kimi K3 on July 16, 2026, calling it the first open model at the 3T class. The Hugging Face model card lists 2.8T total and 104B activated parameters, a 1,048,576-token context, image and text input, and new Kimi Delta Attention and Attention Residuals components.
Moonshot says K3 performs at a frontier level but still trails the top proprietary models overall. It launched on Kimi.com, Kimi Work, Kimi Code, mobile apps, and the Kimi API as kimi-k3, with weights promised by July 27, 2026 under the Kimi K3 License.
Kimi K3 benchmarks and pricing
Values are copied from the linked source on the date shown. They are not IA Compa measurements.
- Output speedas of · Source for Output speed: artificialanalysis.ai (opens in a new tab)
- 39 tokens/s
- Input price per 1M tokensas of · Source for Input price per 1M tokens: kimi.com (opens in a new tab)
- $3.00
- Output price per 1M tokensas of · Source for Output price per 1M tokens: kimi.com (opens in a new tab)
- $15.00
- Context windowas of · Source for Context window: huggingface.co (opens in a new tab)
- 1M tokens
How it compares
Other models in the same category that we have covered, using each model's latest recorded value. (Models)
Kimi K3 reasoning effort
Artificial Analysis tests this model at several reasoning-effort settings. Higher effort usually scores higher but costs more and answers more slowly.
Kimi K3 by effort level
Higher effort lets the model think longer. Here is what each setting scores, how fast it writes, and what it costs on Artificial Analysis.
| Effort | Intelligence Index | Output speed (tokens per second) | Cost to run the index | Blended price (3:1) (per 1M tokens) | Price in / out (per 1M tokens) | First token |
|---|---|---|---|---|---|---|
| Max (Kimi K3 (max)) | n/a | 39 | n/a | $6 | $3 / $15 | 7 s |
| Low (Kimi K3 (low)) · Best: highest effort with a score, used in the leaderboard | 17 | 36 | $1,764 | $6 | $3 / $15 | 9.4 s |
Benchmarked by AA: max · low of 3 available
Where to compare Kimi K3
Independent leaderboards and official resources for comparing this model yourself. Rankings change often; each site shows its own current data.
Leaderboards
General leaderboards
Why Kimi K3 matters for businesses
API pricing is $0.30 per 1M cache-hit input tokens, $3.00 cache-miss input, and $15.00 output. Artificial Analysis measured about 39 output tokens per second and now recommends the newer Kimi K3.1 Pro. For self-hosting, the custom license terms should be reviewed before commercial use.
Sources for this story
- Kimi K3: Open Frontier Intelligence (archived July 16, 2026) (opens in a new tab)Official
Moonshot AI via Internet Archive · web.archive.org
- Kimi K3 (opens in a new tab)Official
Moonshot AI (Kimi blog) · kimi.com
- Kimi K3 model card (opens in a new tab)Official
Moonshot AI on Hugging Face · huggingface.co
- Kimi K3 model analysis (opens in a new tab)Benchmark
Artificial Analysis · artificialanalysis.ai
Last verified: . IA Compa summarizes public announcements for business readers. We are not affiliated with the companies mentioned. Always confirm details, pricing, and availability with the official source.
- #llm
- #moonshot
- #kimi
- #open-weights
- #long-context