TL;DR
Moonshot AI released Kimi K3 on July 16 with API prices matching Claude Sonnet 5’s list rate and independent benchmark results near the frontier. The release could shift competition in China’s AI market from price discounts toward capability, although its promised weights, licence and technical report remain unavailable.
Moonshot AI released Kimi K3 on July 16, pricing the model at $3 per million input tokens and $15 per million output tokens as independent testing placed it close to leading Western systems. The combination challenges the established market view that Chinese AI models compete mainly through lower prices rather than comparable capability.
Kimi 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 per token. It accepts text, images and video and advertises a maximum context window of 1,048,576 tokens, although some service tiers are capped below that level.
Artificial Analysis scored the tested open-weight configuration at 57.1 on its Intelligence Index v4.1, placing it 2.8 points behind the leading model in the supplied comparison. Its long-horizon tracker gave K3 an Elo score of 1,547, a reported 732-point increase from K2.6. Those figures are independent of Moonshot, but they were only one day old when the source report was published and may change as testing expands.
The commercial signal is the price. K3’s $3 input and $15 output rates are about five times the approximate K2-family pricing cited by Thorsten Meyer AI and match the listed rate attributed to Claude Sonnet 5. Sonnet 5’s introductory rate of $2 and $10 through August 31 would temporarily make K3 50% more expensive on those 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.
Price Competition Gives Way to Capability
Chinese developers have gained international users partly through lower API prices, downloadable weights and permissive releases. Moonshot’s decision to charge at a Western mid-tier rate suggests that it expects customers to judge K3 on performance rather than discounting. If buyers accept that price, rival Chinese laboratories may face pressure to invest more heavily in capability and product quality.
The release also puts pressure on Western providers. A Chinese model offered at a similar price, with weights scheduled for release, could appeal to organizations seeking more deployment control than closed systems provide. That competitive effect depends on the licence, hardware requirements and actual performance across business workloads, none of which is fully established.
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.
China’s Low-Cost Model Strategy Shifts
For roughly two years, Chinese AI systems were often positioned as capable, cheaper alternatives to models from major US companies. Moonshot’s earlier K2 family followed that pattern, while other Chinese developers used open or open-weight releases to encourage adoption and local deployment.
K3 changes that calculation through both scale and pricing. Moonshot calls it its most capable model to date, while the source report describes its announced 2.8 trillion parameters as the largest total for an open-weight model. The model remains sparse, however, and total parameters do not equal active computing demand; Moonshot has not disclosed how many parameters are active for each token.
“Our most capable model to date, with 2.8 trillion parameters.”
— Moonshot AI, in launch language cited by Thorsten Meyer AI
large language model for text and video
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Licence and Compute Demands Remain Unknown
K3 cannot yet be independently described as an available open-weight model. Moonshot has promised the weights by July 27, but the licence and technical report were unpublished at the time of the source report. Until those materials appear, users cannot confirm whether commercial use, modification or redistribution will face restrictions.
Moonshot has also not disclosed the active parameter count, detailed training resources or practical infrastructure requirements. The source material discusses export controls and distillation allegations involving Chinese laboratories, but it does not establish that K3 violated rules or relied on improper training methods. The effect of K3’s launch on market share, revenue and customer switching is also unknown.
AI capability benchmarking tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
July 27 Release Will Test Moonshot
Attention now turns to July 27, when Moonshot says it will publish K3’s weights. Developers will examine the licence, model files and hardware demands before deciding whether K3 offers a workable alternative to closed APIs.
Independent laboratories are also expected to run broader evaluations covering coding, reasoning, multimodal tasks, long-context reliability and operating costs. Those results, along with customer response to the $3 and $15 pricing, will show whether K3 can convert benchmark performance into sustained competitive pressure.
AI model pricing comparison
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 latest multimodal model, released through its app, Playground and API. Moonshot says it has 2.8 trillion total parameters in a sparse mixture-of-experts architecture.
How much does Kimi K3 cost?
The listed API rate is $3 per million input tokens and $15 per million output tokens, with cached input priced at $0.30 per million tokens.
Has Kimi K3 matched leading Western models?
Independent Artificial Analysis testing placed K3 near, but not at, the frontier. Its reported score was 2.8 points behind the leader in the cited comparison, and further testing is needed across real workloads.
Is Kimi K3 open source?
Not yet. Moonshot has promised downloadable weights by July 27, but the licence was still unpublished. Open weights do not automatically mean unrestricted open-source use.
How could K3 affect China’s AI market?
K3 may push competition toward capability at comparable prices, weakening the assumption that Chinese models must undercut Western providers. Its actual effect will depend on customer adoption, licensing and operating costs.
Source: Thorsten Meyer AI
