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 price. Independent testing places the model near leading Western systems, but its weights, license, technical report and active parameter count remain unavailable.
Moonshot AI released Kimi K3 on July 16 with near-frontier independent test results and API prices of $3 per million input tokens and $15 per million output tokens, matching Claude Sonnet 5’s listed rate. The launch challenges the view that Chinese AI developers compete mainly through discounts, although claims that K3 has ended China’s price competition remain an interpretation rather than a confirmed market outcome.
Kimi K3 scored 57.1 on Artificial Analysis’ Intelligence Index v4.1, according to figures cited by Thorsten Meyer AI. That placed the tested open-weight configuration 2.8 points behind the leading model in the supplied comparison and close to GPT-5.6 Sol Max. Artificial Analysis also recorded a 732-point Elo increase over K2.6 on its long-horizon tracker, bringing K3 to 1,547.
Moonshot describes K3 as its most capable model to date, with 2.8 trillion total parameters and a sparse mixture-of-experts architecture that routes 16 of 896 experts per token, alongside shared experts. Its published specifications include a maximum 1,048,576-token context window and native text, image and video input. At launch, reasoning is available only at the Max setting.
The commercial change is equally direct. K3’s listed API price is about five times the estimated K2-family rate and, according to Thorsten Meyer AI, makes it the highest-priced model released by a Chinese laboratory. Anthropic’s temporary introductory price for Sonnet 5 is $2 per million input tokens and $10 per million output tokens through August 31, leaving K3 50% more expensive during that offer.
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.
Capability Replaces the China Discount
Chinese AI systems have often been marketed as lower-cost alternatives to models from US developers. K3’s pricing suggests Moonshot believes its model can compete on performance rather than price alone. That could pressure buyers to compare reasoning quality, latency, context limits, deployment options and license terms instead of assuming Chinese models will carry a large discount.
The release also arrived earlier than some observers expected. Thorsten Meyer AI said analysts had anticipated this performance tier in early 2027, making the July 2026 result roughly six months ahead of that forecast. The evidence supports a narrowing performance gap, but one launch cannot establish that a broader price war has stopped across China’s crowded AI market.
AI language model API
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
From K2 Pricing to K3 Scale
Moonshot’s K2 family was associated with an estimated price near $0.60 per million input tokens and $3 per million output tokens. K3 raises those rates to $3 and $15 while expanding the announced model from about one trillion to 2.8 trillion total parameters.
The model’s scale complicates claims that hardware restrictions have forced Chinese laboratories to advance mainly through efficiency. K3 is much larger by total parameter count, but its active parameter count is undisclosed. Because it uses sparse expert routing, total parameters do not reveal inference compute or the hardware needed for each request. The launch alone cannot show how export controls affected development, training costs or access to chips.
“Our most capable model to date, with 2.8 trillion parameters.”
— Moonshot AI, in launch materials cited by Thorsten Meyer AI
large language model development kit
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
License and Compute Details Missing
K3 is being described as an open-weight model, but its weights were not available at publication and Moonshot said they would arrive by July 27. The license has not been published, so developers cannot yet verify whether commercial use, modification or redistribution will be broadly permitted.
Moonshot also has not released the technical report or active parameter count. Its one-million-token context figure is a maximum specification, and access may vary by product tier; the supplied material says Moderato is capped at 256,000 tokens. Independent results are encouraging but were only one day old, while Moonshot’s own benchmark figures remain self-reported. Real-world reliability, operating cost and performance across different workloads require more testing.
AI model training software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
July 27 Release Faces Scrutiny
Attention now shifts to July 27, when Moonshot has promised to publish K3’s weights. Developers will examine the license, model files and deployment requirements, while independent testers are likely to rerun evaluations across reasoning, coding, multimodal tasks and long-context use.
Commercial adoption will provide a separate test of Moonshot’s strategy. If customers accept Western mid-tier pricing, other Chinese laboratories may have more room to raise prices for their strongest systems. If demand remains concentrated on cheaper models, discount competition may continue despite K3’s launch.
AI model evaluation tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
When was Kimi K3 released?
Kimi K3 was released July 16, 2026, and is available through the Kimi app, Playground and API. Moonshot says the model weights will follow on July 27.
How much does Kimi K3 cost?
The listed API rate is $3 per million input tokens and $15 per million output tokens. Cached input is listed at $0.30 per million tokens.
Has Kimi K3 ended China’s AI price war?
No sector-wide end has been confirmed. K3 shows that Moonshot is willing to charge Claude-level prices for one flagship model, but competing Chinese developers may continue offering lower-priced systems.
Is Kimi K3 already open source?
No downloadable weights or published license were available at the time covered by the source material. Until the promised July 27 release, claims about its openness rest on Moonshot’s announcement.
How close is K3 to leading AI models?
Artificial Analysis gave the tested configuration a score of 57.1, which was 2.8 points below the leader in the cited index. Rankings can change as testing expands and models receive updates.
Source: Thorsten Meyer AI
