Kimi K3

Moonshot AI's flagship successor to the K2 line, live on the Kimi API from July 16, 2026 with full open weights following on July 27. A 2.8-trillion-parameter sparse mixture-of-experts model, making it the largest openly released model to date, with MXFP4 weight quantization bringing the download to roughly 1.4TB. Handles text, images and video natively across a 1M-token context. Flat pay-as-you-go pricing with no long-context surcharge, plus a 90% cache-hit discount ($0.30 per 1M cached input tokens).

Kimi K3 strengths

• Largest open-weight model released to date (2.8T params, sparse MoE)
• Native text, image and video input
• 1M-token context with no long-context surcharge
• 90% cache-hit discount suits coding agents
• Open weights allow self-hosting and fine-tuning

Pricing & context

Cost guide: a typical call of about 10K input + 2K output tokens costs roughly $0.060 at list prices. Worth modelling against cheaper tiers before committing high-volume traffic.

When to choose Kimi K3

Kimi K3 is best for teams that want frontier-adjacent, multimodal capability with the option to self-host rather than stay on a closed API. If your workload is more cost-sensitive, weigh it against gpt-oss-120b (≈$0.03 input /1M) first.

Kimi K3 FAQ

How much does Kimi K3 cost?

Kimi K3 is priced at $3.00 per 1M input tokens and $15.00 per 1M output tokens (public API list price), with a 1,048,576 tokens context window. A typical call of about 10K input and 2K output tokens costs roughly $0.060.

What is Kimi K3 best for?

Kimi K3 by Moonshot AI is best for teams that want frontier-adjacent, multimodal capability with the option to self-host rather than stay on a closed API.

How does Kimi K3 pricing compare to GPT-5.4?

Kimi K3 input costs $3.00 per 1M tokens versus $2.50 for GPT-5.4, roughly 1.2x more expensive on input. Output is $15.00 vs $15.00.

Is Kimi K3 multimodal?

Kimi K3 supports text, image, video.