Kimi K3 has 2.8 trillion total parameters in a mixture-of-experts architecture with 896 experts, of which 16 activate per token. That makes it roughly 75% larger than DeepSeek V4 Pro and one of the largest open-weight models released so far.

Is Kimi K3 open weight?

Yes. Moonshot AI announced Kimi K3 on July 16, 2026, with full weights following on July 27, 2026, allowing self-hosting and fine-tuning rather than only hosted API access.

How much does Kimi K3 cost through the API?

Moonshot prices Kimi K3 at $3 per million input tokens on a cache miss, $0.30 per million cached input tokens, and $15 per million output tokens, well below Claude Fable 5's $50 per million output tokens.

How does Kimi K3 rank against other frontier models?

It places second on the Vals AI Index and third on the Artificial Analysis Intelligence Index, behind Claude Fable 5 and GPT-5.6 Sol Max, and first on the Frontend Code Arena benchmark. Moonshot says it beats Claude Opus 4.8 and GPT-5.5 on coding and agent tasks.

Kimi K3: Moonshot AI's 2.8-Trillion-Parameter Open Model, Explained

Moonshot AI's Kimi K3 is a 2.8-trillion-parameter open-weight model that ranks near the top of several benchmarks at a fraction of frontier pricing.

By Quantara Editorial Team · AI Tools · Published Jul 25, 2026 · 9 min read

Moonshot AI announced Kimi K3 on July 16, 2026, with full model weights following on July 27. At 2.8 trillion parameters, it is roughly 75% larger than DeepSeek V4 Pro and is one of the largest open-weight models released to date.

What Is Kimi K3?

Kimi K3 is a mixture-of-experts model with 896 experts, of which 16 are active per token. It supports a one-million-token context window and native multimodal input across text, image, and video, with an always-on "thinking mode" for extended reasoning. Moonshot built the architecture around what it calls Kimi Delta Attention, Attention Residuals, and Stable LatentMoE, which the company says delivers roughly 2.5 times the scaling efficiency of its predecessor, Kimi K2.

Kimi K3 Pricing

Moonshot lists Kimi K3 at $3 per million input tokens on a cache miss, $0.30 per million tokens for cached input (a 90% discount), and $15 per million output tokens. That output price is less than a third of Claude Fable 5's $50 per million output tokens.

Where Kimi K3 Ranks

Kimi K3 places second on the Vals AI Index and third on the Artificial Analysis Intelligence Index, behind Claude Fable 5 and GPT-5.6 Sol Max. It ranks first on the Frontend Code Arena benchmark. Moonshot says Kimi K3 beats Claude Opus 4.8 and GPT-5.5 on coding and agent benchmarks, while trailing Fable 5 and GPT-5.6 Sol on overall intelligence rankings.

Why Moonshot Built It This Way

Moonshot AI president Yutong Zhang described the constraint behind the design: "We knew we didn't have the luxury to simply scale up compute." Rather than compete purely on training-cluster size, Moonshot focused on architectural efficiency to close the gap with better-funded US labs on a smaller compute budget.

Moonshot's Business Position

Moonshot AI raised $2 billion in May 2026 at a valuation above $20 billion, with backers reported to include Alibaba, Tencent, and Meituan. The company's annual recurring revenue is reported above $200 million, and it is reportedly preparing a Hong Kong IPO.

Should You Use Kimi K3?

Teams that want open weights they can self-host or fine-tune.
Coding-heavy workloads, given its first-place Frontend Code Arena result.
Cost-sensitive teams, given output pricing well below Fable 5 and cached-input pricing well below most closed models.
Projects that need a one-million-token context window without a closed-model subscription.

Build Portable Prompts for Kimi K3

Use Verbito's AI Prompt Generator to write model-neutral prompts with clear roles, context, and output formats that transfer cleanly between Kimi K3, Opus 5, and other models.

Sources

Frequently asked questions

How big is Kimi K3?: Kimi K3 has 2.8 trillion total parameters in a mixture-of-experts architecture with 896 experts, of which 16 activate per token. That makes it roughly 75% larger than DeepSeek V4 Pro and one of the largest open-weight models released so far.
Is Kimi K3 open weight?: Yes. Moonshot AI announced Kimi K3 on July 16, 2026, with full weights following on July 27, 2026, allowing self-hosting and fine-tuning rather than only hosted API access.
How much does Kimi K3 cost through the API?: Moonshot prices Kimi K3 at $3 per million input tokens on a cache miss, $0.30 per million cached input tokens, and $15 per million output tokens, well below Claude Fable 5's $50 per million output tokens.
How does Kimi K3 rank against other frontier models?: It places second on the Vals AI Index and third on the Artificial Analysis Intelligence Index, behind Claude Fable 5 and GPT-5.6 Sol Max, and first on the Frontend Code Arena benchmark. Moonshot says it beats Claude Opus 4.8 and GPT-5.5 on coding and agent tasks.