TL;DR
Moonshot AI released Kimi K3 on July 16 at $3 per million input tokens and $15 per million output tokens, matching Claude Sonnet 5’s list price. Independent testing placed K3 close to leading models, but its promised weights, licence and technical report remain unavailable.
Moonshot AI released Kimi K3 on July 16 at $3 per million input tokens and $15 per million output tokens, putting the Chinese model at the same list price as Anthropic’s Claude Sonnet 5. The pricing, paired with an independent benchmark score close to the tested frontier, challenges the market assumption that Chinese laboratories must compete mainly by offering cheaper AI.
K3 is available through the Kimi app, Playground and API. Moonshot describes it as a 2.8-trillion-parameter sparse mixture-of-experts model that routes 16 of 896 experts for each token. It supports text, image and video input and lists a maximum context window of 1,048,576 tokens, although the Moderato service tier is capped at 256,000.
Artificial Analysis scored K3 at 57.1 on its Intelligence Index v4.1, behind Claude Fable 5 at 59.9 and GPT-5.6 Sol Max at 58.9 in the cited comparison. Its long-horizon Elo rating reached 1,547, a 732-point gain over K2.6, while K3 ranked first on Design Arena. These are independent results, but they reflected only the configurations tested and were one day old when the source analysis was published.
The commercial shift is just as pronounced. K3 costs roughly five times more than the K2 family, based on the cited approximate pricing, and Thorsten Meyer AI described it as the most expensive model released by a Chinese laboratory. Anthropic’s temporary Sonnet 5 introductory price of $2 for input and $10 for output means K3 is currently 50% more expensive on those rates through August 31.
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.
China’s AI Discount Narrows
Chinese model providers have gained users partly through low API prices, downloadable weights and capable performance. K3 changes that proposition: Moonshot is asking customers to compare it with Western models on capability rather than price alone. If buyers accept the rate, other Chinese laboratories may have more room to raise prices and fund costly training and inference.
The result also puts pressure on Western vendors. K3 sits 2.8 points behind the highest cited score and was released months before analysts cited by Thorsten Meyer AI expected a Chinese model at this tier. That does not establish parity across every task, but it makes benchmark leadership, product reliability and model access more central competitive factors.
Designing Large Language Model Applications: A Holistic Approach to LLMs
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
From K2 Pricing to K3 Scale
Moonshot’s earlier K2 models fit the broader image of Chinese AI as a lower-cost alternative. K3 reverses that positioning while expanding total parameter count from about 1 trillion to 2.8 trillion, above the cited 1.6 trillion for DeepSeek V4-Pro.
The scale complicates claims that export controls forced Chinese developers to rely chiefly on efficiency. K3’s existence shows that Moonshot can announce a model at this size, but total parameters do not equal computing demand in a sparse architecture. Without the active parameter count or technical report, the release cannot by itself establish how much hardware was used or whether particular controls failed.
“Our most capable model to date, with 2.8 trillion parameters.”
— Moonshot AI launch materials
![Claude AI for Beginners Bible: [5 in 1] The Ultimate Guide to Automate Your Work, Save Hours Every Week, and Use AI for Real-World Results](https://m.media-amazon.com/images/I/415+fSJacsL._SL500_.jpg)
Claude AI for Beginners Bible: [5 in 1] The Ultimate Guide to Automate Your Work, Save Hours Every Week, and Use AI for Real-World Results
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Licence and Weights Still Missing
K3 has been described as open-weight, but its weights were not available at launch. Moonshot has promised publication by July 27, while the licence and technical report remain unpublished. Until those materials appear, claims that K3 is the largest open-source model cannot be verified.
Moonshot has also not disclosed the active parameter count, a key measure for interpreting the computing requirements of a sparse model. Only the Max reasoning setting was available at launch, and it is unclear when other reasoning levels will arrive. The company’s own benchmark results are self-reported, while the independent Artificial Analysis figures cover a limited and changing test set.
Claims that K3 proves export controls ineffective also remain interpretation rather than established fact. Public information does not show the model’s hardware supply, training cost or development constraints. Separate accusations by Anthropic that Chinese laboratories used improper distillation are disputed claims and do not establish how K3 was trained.
The Applied AI Universe Coding Guide: Adversarial Attacks: A Hands-On Handbook for Attacking and Measuring Every AI Model (The Adaptive AI Codex Series)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
July 27 Release Test
Attention now shifts to July 27, when Moonshot says it will release K3’s weights. Researchers and customers will then look for the licence, technical report and active parameter count, test whether independent results hold across more workloads, and compare the model’s operating costs with Claude Sonnet 5 after Anthropic’s introductory pricing ends.
As an affiliate, we earn on qualifying purchases.
Key Questions
When was Kimi K3 released?
Moonshot AI released Kimi K3 on July 16, 2026. It is available through the Kimi app, Playground and API.
How much does Kimi K3 cost?
The listed API price is $3 per million input tokens and $15 per million output tokens, with cached input priced at $0.30 per million tokens.
Is Kimi K3 better than leading Western models?
Independent testing placed K3 close to the leading tested models, but not at the top of the cited Intelligence Index. Results vary by benchmark, configuration and task, so broad superiority is not confirmed.
Is Kimi K3 open source?
Not yet in a verifiable sense. Moonshot has promised weights by July 27, but the licence was unpublished at launch, preventing a full evaluation of reuse rights.
Has K3 ended AI price competition in China?
No market-wide outcome has been established. K3 shows that Moonshot is no longer relying on a large price discount for its flagship model, but other Chinese providers may continue competing through lower prices or permissive licences.
Source: Thorsten Meyer AI
