Chinese AI labs · Vision & multimodal
Kimi K3
Frontier long-context agent work where you can hit cache: the 10x cache-hit discount ($0.30 vs $3.00) matters far more than the headline rate, and $15 output makes it expensive for chatty workloads.
Open weights · Community or custom license · Data checked September 6, 2026 · Not a hands-on evaluation
Model ID kimi-k3
- Parameters
- 2.8T total · 104B active2.8T total / 104B active MoE (896 experts, 16 active)
- Architecture
- Mixture of experts
- Context window
- 1,048,576 tokensMaximum input
- Max output
- Not recordedPer response
- Input price
- $2.98 per 1M tokens
- Output price
- $14.90 per 1M tokens
- Cached input
- $0.30 per 1M tokens
- Blended price
- $5.96 per 1M tokens3:1 input to output
- Knowledge cutoff
- Not recorded
- Released
- 2026-06
- Status
- gaFlagship in its lineup
- Size band
- 120B and above
License and openness
Kimi K3 is released under the Kimi K3 License (custom, license_name kimi-k3 on Hugging Face). This is a community or custom license with its own conditions, which can include acceptable-use rules, user thresholds or naming requirements. Read it before commercial use. License terms can change between versions; confirm the text published with the weights you download.
Capabilities
| Input modalities | text, image, video |
|---|---|
| Output modalities | text |
What it is for
At a three-to-one input-to-output ratio, Kimi K3 costs $5.96 per million tokens blended. A workload of one million input and 330,000 output tokens per day would run about $236.91 per month at list price, before caching or batch discounts.
Cost at list price
| Input | $2.98 per 1M tokens |
|---|---|
| Output | $14.90 per 1M tokens |
| Cached input | $0.30 per 1M tokens |
| Blended, 3:1 | $5.96 per 1M tokens |
Sources
- Source: platform.kimi.com ↗
- Chinese AI labs pricing ↗
- API docs ↗
- Full reference record for Kimi K3 ↗
- Chinese AI labs in the reference ↗
Sources checked September 6, 2026. Specifications and prices change without notice; confirm against the provider before you commit.
Other open-weight models from Chinese AI labs
- Kimi K2.7 Code · 1T total · 32B active
- Kimi K2.7 Code Highspeed · 1T total · 32B active
- Kimi K2.6 · 1T total · 32B active
- GLM-5.3 · 753B
- GLM-5.3-Flash · 320B total · 18B active
- GLM-5.2 · 753B
- GLM-5 · 754B
- GLM-4.7 · Not recorded
- GLM-4.5-Air · Not recorded
- MiniMax-M3 · 427B total · 23B active
- MiniMax-M2.7 · 229B
- MiniMax-M2.7-highspeed · 229B