TL;DR
Moonshot AI released Kimi K3 on July 16, pricing it at $3 per million input tokens and $15 per million output tokens, matching Claude Sonnet 5’s list price. Independent benchmark results place K3 close to leading models, but its weights, licence, technical report and active parameter count remain unavailable.
Moonshot AI released Kimi K3 on July 16, placing the Chinese model close to leading systems on an independent benchmark while charging $3 per million input tokens and $15 per million output tokens. The price matches Claude Sonnet 5’s standard rate, signaling that Moonshot is competing on capability rather than maintaining the large discount commonly associated with Chinese AI models.
Kimi K3 recorded a score of 57.1 on Artificial Analysis Intelligence Index v4.1, according to results cited by Thorsten Meyer AI. That left it 2.8 points behind the highest tested score and placed its tested configuration fourth. It also reached 1,547 on the evaluator’s long-horizon tracker, a 732-point Elo increase over Kimi K2.6.
Moonshot describes K3 as a 2.8-trillion-parameter sparse mixture-of-experts model, routing 16 of 896 experts for each token alongside shared experts. The company lists a 1,048,576-token maximum context window, support for text, image and video input, and adjustable reasoning effort, although only the Max reasoning setting was available at launch.
The model is live through the Kimi app, Playground and API. Moonshot’s pricing is roughly five times the cited rate for the K2 family and matches Claude Sonnet 5’s $3 input and $15 output list prices. Claude’s temporary introductory rate of $2 and $10 through August 31 means K3 currently costs 50% more at the listed API rates.
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.
K3 Abandons China’s AI Discount
The pricing changes the competitive signal from Moonshot. Chinese developers have often attracted users with lower API costs, downloadable weights or both. By charging the same list price as Sonnet 5, Moonshot is making a commercial claim that K3 can compete without a large discount.
That does not mean China’s AI price competition has ended across the market. It does show that one prominent Chinese laboratory has moved away from discount-led positioning for its flagship model. Buyers comparing providers may now place greater weight on reasoning quality, reliability, deployment rights and ecosystem support rather than token price alone.
Vision-Language Models in Production: Architecting Multimodal LLM Applications: From Vision-Language API to Self-Hosted Model (Production AI Engineering Series)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
A Faster Climb Toward Frontier Models
K3 arrives after two years in which Chinese laboratories were widely framed as offering capable models at lower prices. The cited benchmark results suggest Moonshot reached this performance tier months earlier than some analyst expectations, although forecasts differ and benchmark rankings can change quickly.
The model’s scale also complicates claims that export restrictions forced Chinese developers to rely mainly on efficiency. K3’s 2.8 trillion total parameters exceed Moonshot’s previous K2 family and other large announced open-weight systems. Because K3 uses a sparse expert architecture, however, total parameter count does not directly show its compute use per token.
“Our most capable model to date, with 2.8 trillion parameters.”
— Moonshot AI launch materials
Weights and Licence Still Missing
Despite descriptions of K3 as open-weight, its weights were not available at launch. Moonshot said they would arrive by July 27, while the licence and technical report remained unpublished. Until those materials appear, claims about openness, commercial reuse and modification rights cannot be independently checked.
Moonshot has also not disclosed the model’s active parameter count, making efficiency comparisons incomplete. The one-million-token context figure is a stated maximum rather than a guarantee for every service tier, and real-world performance outside selected evaluations remains uncertain. Moonshot’s benchmark claims are self-reported; the Artificial Analysis results are independent but represent an early test snapshot.
July 27 Release Becomes Key Test
Attention now shifts to Moonshot’s promised July 27 weights release. Developers will examine the licence, hardware requirements, active parameters and whether independent deployments reproduce the reported performance. Longer testing will also show whether K3’s quality supports its Sonnet-level price and whether other Chinese laboratories follow Moonshot toward higher flagship pricing.
Key Questions
When was Kimi K3 released?
Kimi K3 was released on July 16, 2026, through the Kimi app, Playground and API. Its downloadable weights are promised for July 27.
How much does Kimi K3 cost?
Moonshot lists K3 at $3 per million input tokens and $15 per million output tokens. That matches Claude Sonnet 5’s standard list price, though Claude has a lower introductory rate through August 31.
Is Kimi K3 the leading AI model?
No. The cited Artificial Analysis test gave K3 a 57.1 score, leaving it 2.8 points below the top tested result. It is close to the frontier but did not lead the overall index.
Is Kimi K3 already open-weight?
Not yet. Moonshot has announced an open-weight release for July 27, but the weights and licence were unavailable at launch. Its reuse rights remain unconfirmed until the licence is published.
Does K3 mean China’s AI price war is over?
No market-wide conclusion is confirmed. K3 shows that Moonshot is willing to charge Western mid-tier prices for one flagship model, but other Chinese providers may continue competing through lower prices or permissive releases.
Source: Thorsten Meyer AI
