Kimi K3
A 2.8-trillion-parameter open model that developers prefer for front-end work.
Overall Intelligence
#17
of 56
6.9 pts behind #1
0
races led
10
weeks at #1
81.8
score
- Released
- Jul 16, 2026
- 3 mo ago
- Context
- 1M
- 128K max output
- Price
- $3 / $12
- per 1M in/out · $5.25 blended
- Speed
- 70 tok/s
- 3s to first token
- Modalities
- text · image
- outputs text
- Weights
- Open
- 2.8T MoE (open)
The 30-second version
Moonshot's 2.8-trillion-parameter open model. Kimi K3 is a favourite with developers for front-end and UI work, and it is the strongest open-weights model in several coding races.
- 1Leads our open-weights field on front-end and UI generation.
- 2Open weights, but its size makes self-hosting a serious infrastructure project.
- 3Agentic coding close to the closed mid-tier at $3 / $12 via API.
- 4Weaker than closed leaders on calibration and factuality.
What you need to know
Purpose
Push open models to the frontier on coding and agents, with a particular eye for polished front-end output.
What Moonshot optimised for
Scale and agentic post-training, with heavy emphasis on web and UI generation quality.
Unique strengths
- Beautiful, idiomatic front-end code.
- Strong long-context agentic coding.
- Open weights at frontier scale.
Where it ranks
Current snapshot · illustrative scores · click any race to watch it
Kimi K3Best on each axis
Overall
Coding
Hard Reasoning
Math
Science
Writing & Language
Multimodal
Tool Use & Agents
Speed & Cost
Overall Intelligence over time
Best use cases
01
Landing pages and UI
Turning a description or screenshot into polished components.
02
Open-stack coding agents
Teams that want frontier-class coding without a closed vendor.
03
Long-context code review
Reading large front-end codebases for consistency issues.
Strengths
Front-end
Leads open models on our frontend race and challenges the closed leaders.
Openness
Weights available for audit and fine-tuning.
Weaknesses
Hosting cost
Running 2.8T parameters yourself is expensive.
Calibration
More confident than it should be on uncertain facts.
Prompts that play to its strengths
Screenshot to component
Recreate this pricing section as a responsive React component with Tailwind. Match spacing and hierarchy closely, use semantic HTML, and make the annual/monthly toggle accessible. <screenshot>
Why it works: Plays to its visual taste and front-end fluency.
When to choose it, and when not to
Pick something else when…
- Fact-heavy research where calibration matters.
- Small-footprint deployments.
Update history
The distilled version first; open any release for the deeper breakdown.
The Kimi family