TL;DR
Moonshot AI released Kimi K3 on July 16, pricing it at $3 per million input tokens and $15 per million output tokens. Independent testing places it near leading Western models, but its promised weights, license and technical report remain unavailable.
Moonshot AI released Kimi K3 on July 16, introducing a 2.8-trillion-parameter model priced at $3 per million input tokens and $15 per million output tokens. Its near-frontier independent test result and parity with Claude Sonnet 5’s list price mark a change from the lower-cost positioning associated with many Chinese AI models.
Kimi K3 is available through the Kimi app, Playground and API. Moonshot says the model accepts text, images and video, supports a maximum context window of 1,048,576 tokens and uses a sparse mixture-of-experts architecture that routes 16 of 896 experts for each token.
Artificial Analysis placed K3 at 57.1 on its Intelligence Index v4.1, compared with 59.9 for Claude Fable 5 and 58.9 for GPT-5.6 Sol Max. That leaves K3 2.8 points below the tested leader. The independent tracker also recorded an Elo score of 1,547, a 732-point increase over K2.6, while ranking K3 first on its Design Arena.
The pricing is about five times that attributed to the K2 family and matches Claude Sonnet 5’s standard rate. Anthropic’s temporary Sonnet 5 introductory price of $2 per million input tokens and $10 per million output tokens runs through August 31, leaving K3 more expensive during that period.
Kimi K3: the gap closed six months early — and China stopped competing on price
Every write-up today says “China caught up.” True — and the less interesting half. The other half: K3 costs 5× its predecessor, making it the most expensive Chinese model ever, priced at exact parity with Claude Sonnet 5. A benchmark is a claim. A price is a claim the vendor has to live with.
For two years the thesis was “cheap alternative.” Moonshot just abandoned it. Vendors discount when they’re compensating for something — Moonshot has stopped compensating. With Sonnet 5’s intro rate at $2/$10 through 31 Aug, K3 currently costs 50% more than the model it’s priced against. The competition just moved from cheap vs good to good vs good at the same price, with one of them open — and you can’t answer that with a discount.
The story we’ve told: export controls forced Chinese labs into efficiency. But K3 is 2.8T — the largest open model ever, ~3× K2, vs DeepSeek V4-Pro’s 1.6T. That’s not more with less. That’s more with more. Caveat: sparse MoE, active params undisclosed — total ≠ FLOPs. But if the controls were binding at the frontier, this model shouldn’t exist.
Anthropic has accused Moonshot, Z.AI, MiniMax, Alibaba & DeepSeek of “illicit” distillation — possibly well-founded; I can’t assess it. But one day earlier, Thinking Machines said Inkling’s post-training bootstrapped on Kimi K2.5 — reported as ecosystem health. Same verb, different flag, different word. If the distinction is real, someone should articulate it.
Two things changed, neither in the headlines. The discount is gone — anyone whose China strategy was “they’re cheaper” needs a new strategy. And the controls didn’t work — six months early, biggest model ever, from a lab that was supposed to be compute-starved, while Washington’s options narrow to loosening restrictions on its own labs, criminalising distillation, or subsidising American open weights. That’s not containment. It’s a menu of concessions. The gap is 2.8 points and closing. The price is Sonnet’s. The weights are ten days out. Everything that matters happens on 27 July.
Chinese AI Moves Beyond Discounts
The release suggests that Moonshot is testing demand on capability instead of relying on a large price advantage. Chinese developers have often competed through lower API fees, downloadable weights or both; K3 asks customers to pay Western mid-tier prices for performance that independent testing places close to the leading group.
That change could pressure rival laboratories to compete on model quality, deployment flexibility and licensing. If Moonshot publishes usable weights under permissive terms, K3 could pair near-frontier results with customer-controlled deployment. Until the license appears, that advantage remains a possibility rather than a confirmed feature.
The GPT-4 Millionaire: Future of Business Featuring Microsoft 365 Copilot: How to Leverage AI Language Models to Grow Your Company and How AI-driven Language Models Will Revolutionize the Way We Work
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
K3 Accelerates Moonshot’s Model Gains
Moonshot describes K3 as its most capable model to date. The company reports 2.8 trillion total parameters, nearly three times the K2 family’s stated scale, although total parameter count does not show how much computation is used for each request in a sparse expert model.
The Thorsten Meyer AI dispatch said analysts had expected models at this performance tier in early 2027, framing K3 as arriving about six months earlier than forecast. No named analyst consensus or published forecast accompanied that claim, so the timing comparison should be treated as the dispatch’s interpretation, not an independently established deadline.
“Our most capable model to date, with 2.8 trillion parameters.”
— Moonshot AI launch materials
License and Compute Details Missing
K3’s weights were not available at launch, despite descriptions of the model as open-weight or open-source. Moonshot says the weights will arrive by July 27, but the license has not been published, leaving commercial use, modification and redistribution rights unknown.
The company also has not released a technical report or active parameter count. Only the Max reasoning setting is available at launch, and access to the advertised one-million-token context may depend on the service tier. These gaps limit independent checks of efficiency, deployment cost and benchmark reproducibility.
Broader claims that K3 proves export controls have failed are also unconfirmed. Model scale alone does not reveal which hardware was used, when it was obtained or how much computing capacity the training run required.
July 27 Tests Moonshot’s Promise
Attention now turns to Moonshot’s July 27 weight release. Developers will be able to inspect the files, test performance outside Moonshot’s hosted service and determine whether the license supports commercial and modified deployments.
Independent testers are also likely to examine reasoning modes, long-context reliability and operating requirements. Those results will show whether K3’s benchmark position and premium price hold across real workloads, or whether the launch claims need further qualification.
Key Questions
What is Kimi K3?
Kimi K3 is Moonshot AI’s newest multimodal model, with a stated 2.8 trillion total parameters. It accepts text, image and video input and is available through Moonshot’s app, Playground and API.
How much does Kimi K3 cost?
The API costs $3 per million input tokens, $15 per million output tokens and $0.30 per million cached input tokens, according to Moonshot’s published pricing.
Is Kimi K3 open-source?
Not at launch. Moonshot has promised downloadable weights by July 27, but the license and usage terms remain unpublished, so calling the model open-source is premature.
Does Kimi K3 outperform leading Western models?
Current independent results place K3 close to, but below, the tested leaders. Artificial Analysis scored it at 57.1, 2.8 points behind its top listed configuration.
Why is K3’s pricing drawing attention?
K3 matches Claude Sonnet 5’s standard list price, departing from the discount strategy associated with many Chinese models. The pricing indicates that Moonshot believes K3 can compete primarily on capability.
Source: Thorsten Meyer AI
