TL;DR
Moonshot AI released Kimi K3 on July 16, with independent testing placing it close to leading AI models. Its Sonnet-level pricing signals a move away from competing mainly on cost, although its promised weights, license and technical report remain unavailable.
Moonshot AI released Kimi K3 on July 16, with independent testing placing the model 2.8 points behind the measured AI frontier and the company setting prices at $3 per million input tokens and $15 per million output tokens. The combination of near-leading performance and pricing comparable with Anthropic’s Claude Sonnet 5 marks a shift from the low-cost positioning associated with Chinese AI models.
Kimi K3 is available through the Kimi app, Playground and API. Moonshot says the model contains 2.8 trillion total parameters, uses a sparse mixture-of-experts architecture and routes 16 of 896 experts per token. The company also lists a maximum context window of 1,048,576 tokens and support for text, image and video input.
Artificial Analysis scored K3 at 57.1 on its Intelligence Index v4.1, placing the tested configuration fourth overall and the model family effectively third, according to Thorsten Meyer AI’s account of the results. K3 was 2.8 points behind the leading score and 0.54 points behind the cited Sol xhigh configuration. The model also ranked first on Design Arena and reached 1,547 on the evaluator’s long-horizon tracker, a 732-point increase over K2.6.
The pricing is as consequential as the rankings. K3 costs about five times more than the earlier K2 family, based on the prices reported by Thorsten Meyer AI, and matches Claude Sonnet 5’s regular list price. During Anthropic’s introductory rate through August 31, however, K3 is 50% more expensive than Sonnet 5 on both listed input and output 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.
Pricing Shifts the Competitive Test
K3’s release suggests Moonshot believes it can compete on model capability rather than price alone. Thorsten Meyer AI described the price as the larger signal because it exposes the company to direct comparisons with established Western products at the same regular list price. Customers can now judge K3 on reliability, developer tooling and output quality without a large discount shaping the decision.
The independent ranking also narrows the measured distance between a Chinese model and the highest-scoring systems in the cited test. If K3’s results hold across wider use, Western AI providers face another high-performing competitor, while developers may gain access to downloadable model weights. Benchmark proximity does not establish commercial adoption, production reliability or lower operating costs.
AI language model API access
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
China’s Discount Era Recedes
Chinese AI systems have often been presented as less expensive alternatives to closed models from US companies. K3 departs from that pattern: Moonshot is charging Western mid-tier API prices while promising weights that users could later run through other infrastructure.
Moonshot describes K3 as its most capable model to date and says its 2.8 trillion parameters make it larger by total parameter count than previously announced open-weight systems cited in the source material. That figure does not reveal computation per token because K3 is a sparse mixture-of-experts model, and Moonshot has not disclosed its active parameter count. The scale also complicates claims that Chinese laboratories reached the frontier mainly through doing more with limited computing resources.
“Our most capable model to date, with 2.8 trillion parameters.”
— Moonshot AI, in its K3 launch materials
AI model performance testing tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
License and Efficiency Data Missing
K3 cannot yet be independently inspected as a downloadable model. Moonshot has promised weights by July 27, but the license remains unpublished. Until those terms appear, descriptions of K3 as open source or permissively reusable are unconfirmed.
Moonshot has also not published the full technical report or active parameter count. Only the Max reasoning setting was available at launch, while lower reasoning settings had not shipped. The advertised one-million-token window is a maximum specification, and actual limits may differ by service tier. Artificial Analysis results offer independent evidence of performance, but they were only one day old in the supplied reporting and cannot answer questions about uptime, latency or real-world cost.
AI input output token counters
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
July 27 Becomes the Test
Attention now turns to Moonshot’s July 27 deadline for releasing K3’s weights. Developers will examine the license, hardware requirements and reproducibility of the reported performance before deciding whether the model offers a practical open-weight alternative.
Further independent tests should show whether K3’s benchmark position holds across coding, reasoning, media input and long-context tasks. Usage data will also be needed to establish whether its early technical showing becomes commercial adoption.
AI model licensing and weights
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is Kimi K3?
Kimi K3 is Moonshot AI’s newest multimodal model, released on July 16, 2026. Moonshot reports 2.8 trillion total parameters and support for text, image and video input.
Why is Kimi K3 attracting attention?
Independent benchmark testing placed K3 near leading models, while its API price matches Claude Sonnet 5’s regular list price. That pairing challenges the idea that Chinese models compete mainly by being cheaper.
Is Kimi K3 open source now?
No. Moonshot has promised downloadable weights by July 27, but neither the weights nor the governing license was available at the time covered by the source material.
How much does the Kimi K3 API cost?
Moonshot lists K3 at $3 per million input tokens and $15 per million output tokens, with a reported cached-input price of $0.30 per million tokens.
Does K3 prove Moonshot has matched the leading AI models?
No. Artificial Analysis placed K3 close to, but below, the leading score in its cited index. Broader testing and production use are still needed to measure reliability, speed and operating cost.
Source: Thorsten Meyer AI
