TL;DR
Moonshot AI released Kimi K3 on July 16 at $3 per million input tokens and $15 per million output tokens, about five times the K2 family’s estimated pricing. Independent testing places the model close to leading rivals, but its weights, license, technical report and active parameter count remain unavailable.
Moonshot AI released Kimi K3 on July 16 at $3 per million input tokens and $15 per million output tokens, roughly five times the estimated price of its K2 family and equal to Claude Sonnet 5’s standard rate. The launch marks a sharp change in how a leading Chinese AI laboratory is positioning its flagship model: capability rather than a large price discount is now the central sales case.
Kimi K3 is available through the Kimi app, Playground and API. Moonshot describes it as a 2.8-trillion-parameter, highly sparse mixture-of-experts model that routes 16 of 896 experts per token. Its published specifications include a 1,048,576-token maximum context window and native text, image and video input.
Independent testing cited in the source material gives K3 a score of 57.1 on Artificial Analysis Intelligence Index v4.1, 2.8 points behind the leading tested model. It also reportedly reached 1,547 Elo on the evaluator’s long-horizon tracker, a 732-point gain over K2.6, and ranked first on Design Arena. Those results are independent of Moonshot’s self-reported benchmarks, though they reflect testing conducted soon after release.
The pricing change is unusually large. The K2 family cost about $0.60 for input and $3 for output per million tokens, according to the supplied comparison, while K3 charges $3 and $15. That matches Claude Sonnet 5’s list price. During Sonnet 5’s reported introductory offer of $2 and $10 through August 31, however, K3 costs 50% more on both measures.
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 Tests Capability-Led Competition
Chinese AI laboratories have spent two years gaining adoption through a mix of competitive performance, low API prices and downloadable weights. K3 weakens the assumption that this strategy will persist at the highest capability tier. Moonshot is asking customers to pay Western mid-tier rates, indicating that it believes the performance gap is small enough to compete without a broad discount.
That shift could affect developers choosing between Chinese and US model providers, as well as companies whose competitive plans rely on Chinese services remaining cheaper. It does not prove that China’s AI price competition has ended; other laboratories may keep cutting rates. It does show that one major Chinese vendor has broken from the low-price formula for its newest flagship.
Generative AI on AWS: Building Context-Aware Multimodal Reasoning Applications
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
From K2 Discounts to Parity
Kimi K3 follows Moonshot’s lower-priced K2 line and arrives amid rapid gains by Chinese model developers. The supplied Artificial Analysis results place K3 close to Claude Fable 5 and GPT-5.6 Sol Max, while Moonshot calls it its most capable model. Thorsten Meyer AI argues that the combination of higher pricing and a narrow benchmark gap matters more commercially than another claim that China has caught up.
The model’s scale also complicates claims that export restrictions have forced Chinese laboratories to advance mainly through efficiency. At 2.8 trillion total parameters, K3 is about three times the size of K2 and larger than the other announced open-weight systems cited in the source. Because it uses sparse expert routing and Moonshot has not disclosed active parameters, total size does not reveal its per-token computing cost.
“Our most capable model to date, with 2.8 trillion parameters.”
— Moonshot AI, in its K3 launch description
large language model developer tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
License and Compute Details Missing
K3’s model weights were not available at launch; Moonshot has promised them by July 27. The license has not been published, so descriptions of K3 as open source or fully open weight remain premature. Its technical report and active parameter count are also missing, limiting independent scrutiny of training methods, operating costs and efficiency.
Several product limits need clarification. Moonshot advertises a 1,048,576-token maximum context, but service tiers may impose lower caps, and only the Max reasoning setting was available at launch. It is also unclear whether K3’s pricing will hold after the initial release or whether rivals will respond with discounts, new models or revised rates.
artificial intelligence input output tokens
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
July 27 Release Faces Scrutiny
The next test comes on July 27, when Moonshot says it will publish K3’s weights. Developers will then be able to examine whether the model can be deployed outside Moonshot’s hosted services and whether its license permits broad commercial use, modification and redistribution.
Independent evaluators are also expected to expand testing across coding, reasoning, multimodal tasks and long-context workloads. Those results, along with the missing technical report, will help establish whether K3’s performance holds across practical use cases and whether its premium pricing attracts sustained demand.
AI model performance testing software
As an affiliate, we earn on qualifying purchases.
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, while downloadable weights are promised for July 27.
How much does Kimi K3 cost?
K3 costs $3 per million input tokens and $15 per million output tokens. Cached input is listed at $0.30 per million tokens. The standard input and output rates match Claude Sonnet 5’s list pricing.
Has Kimi K3 ended China’s AI price war?
No industry-wide end has been confirmed. K3 shows that Moonshot is willing to price a flagship model without a large Chinese-market discount, but competing laboratories may continue offering cheaper services or cutting rates.
Is Kimi K3 open source?
That description cannot yet be confirmed. The weights and license were unpublished at launch, and Moonshot has promised the weights by July 27. The license will determine whether K3 meets common definitions of open-source or permissive open-weight software.
How close is Kimi K3 to leading AI models?
Artificial Analysis v4.1 reportedly scored K3 at 57.1, placing it 2.8 points behind the leading tested model. The result suggests a narrow benchmark gap, but wider testing is still needed before drawing conclusions about performance across production workloads.
Source: Thorsten Meyer AI
