Overview
Kimi is the assistant and model family from Moonshot AI, available as consumer apps, an API, and — unusually for a frontier model — downloadable weights.
Its current flagship, Kimi K3, reached the API on 16 July 2026 and published its weights on 27 July 2026, making it the first open-weight model in the 3-trillion-parameter class. It is a Mixture-of-Experts model with 2.8 trillion total parameters, 104 billion of which activate per token, a 1,048,576-token context window, native vision covering both images and video, and reasoning that is always on.
The API is both OpenAI- and Anthropic-compatible, so trying it against an existing integration is usually a base URL change. Pricing is $3.00 per million input tokens and $15.00 per million output — the same rate as Claude Sonnet 5, for a model that competes with tiers priced well above it — with cached input at $0.30 per million.
Key Features
- 1,048,576-token context window (1M)
- Open weights under the Kimi K3 License
- Native vision for images and video
- Always-on reasoning with three effort levels
- OpenAI- and Anthropic-compatible API
- Automatic prompt caching at a 10x input discount
- K3 Swarm Max variant for parallel multi-agent workloads
Pricing
Kimi apps
Free tier
Consumer web, mobile, and Kimi Work apps; plan limits change over time.
API
$3 / $15 per 1M tokens
Pay-as-you-go input/output, with cached input at $0.30 per million.
Open weights
No licence fee
Self-host the ~1.4 TB weights; revenue and attribution thresholds apply.
Screenshots
Pros & Cons
👍 Pros
- +Frontier-scale capability at mid-tier pricing
- +Open weights — portability and no hard vendor lock-in
- +1M-token context with a 10x cached-input discount
- +Native video input, not just still images
- +Drop-in compatible with OpenAI and Anthropic SDKs
👎 Cons
- −Reasoning cannot be disabled, so trivial requests carry a cost floor
- −reasoning_effort defaults to the most expensive level
- −No image generation — analysis only
- −Licence is more restrictive than MIT above revenue and user thresholds
- −Not available as a managed model on AWS, Azure, or Google Cloud
Kimi Alternatives
Considering other options? Here are the top alternatives to Kimi.
Frequently Asked Questions
What is Kimi K3?
Moonshot AI's flagship model and the first open-weight model in the 3-trillion-parameter class: 2.8T total parameters with 104B active per token, a 1M-token context window, native vision and video, and always-on reasoning. It launched on the API on 16 July 2026, with weights published on 27 July 2026.
How much does Kimi K3 cost?
$3.00 per million input tokens and $15.00 per million output on the Kimi API, with cached input at $0.30 per million. That matches Claude Sonnet 5's rate and sits 40% below Claude Opus 5. The consumer apps have a free tier.
Is Kimi open source?
Kimi K3 has open weights, published under a custom Kimi K3 License rather than MIT. Commercial use is broadly permitted, but Model-as-a-Service operators above $20M in revenue over any consecutive 12 months need a separate agreement with Moonshot AI.
Related Articles
Kimi K3: Specs, Pricing and How to Use It (2026 Guide)
Kimi K3 explained: 2.8T parameters, 1M-token context, $3/$15 pricing, and open weights. The complete guide plus 34 deep dives.
7 min read
Kimi K3 for Coding: Setup, Settings and Real Limits
Kimi K3 is an open-weight model tuned for long-running code work. The settings, workflow, and failure modes that matter on this model.
4 min read
Kimi K3 for AI Agents: Long-Horizon Runs and Swarms
Kimi K3 was built for agentic work: always-on reasoning, a 1M-token window, and a parallel Swarm Max variant. How to actually use it.
4 min read