TL;DR
Moonshot AI released Kimi K3 on July 16, pairing near-frontier independent benchmark results with API pricing equal to Claude Sonnet 5’s list price. The model is live, but its promised weights, license, technical report and active parameter count have not been published.
Moonshot AI released Kimi K3 on July 16, putting a 2.8-trillion-parameter model into its app, Playground and API with independent benchmark results close to leading systems. Its $3-per-million input-token and $15 output-token pricing, matching Claude Sonnet 5’s stated list price, signals that Moonshot is competing on capability rather than relying on a steep Chinese-model discount.
Moonshot describes K3 as its most capable model to date. The company says it uses a sparse mixture-of-experts architecture that routes 16 of 896 experts per token, alongside shared experts, Kimi Delta Attention and Attention Residuals. It supports text, image and video input and a maximum context window of 1,048,576 tokens.
On the independent Artificial Analysis Intelligence Index v4.1, K3’s tested open-weight configuration scored 57.1, 2.8 points behind the frontier score cited in the source material and 0.54 behind Sol xhigh. Artificial Analysis also recorded a 732-point long-horizon Elo gain over K2.6, bringing K3 to 1547, while the model ranked first on Design Arena.
The release price is about five times the reported K2-family rate. Thorsten Meyer AI characterizes K3 as the most expensive model shipped by a Chinese laboratory, although that comparison depends on the products and pricing tiers included. Claude Sonnet 5 has a temporary introductory rate of $2 for input and $10 for output through August 31, making K3 50% more expensive during that period.
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 Challenges Mid-Tier Pricing
K3 changes the commercial comparison facing developers. Chinese models have often been presented as lower-cost alternatives, but Moonshot is asking customers to pay Western mid-tier prices for this release. The pricing puts greater weight on measured capability, reliability and deployment terms rather than savings alone.
The independent score also places K3 near the tested frontier, supporting the view that the performance gap has narrowed faster than some industry forecasts anticipated. Thorsten Meyer AI says analysts had expected this capability tier in early 2027, making the July release roughly six months earlier than that forecast. The forecast comparison is an interpretation, not a technical finding from Artificial Analysis.
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.
A Six-Month Benchmark Leap
Moonshot’s earlier K2 models were associated with lower API costs and downloadable weights. K3 combines a reported threefold increase in total parameters with a much higher usage price, changing the company’s position from a discount supplier toward a direct competitor to proprietary services.
The model’s scale also complicates claims that export controls forced Chinese laboratories to advance mainly through efficiency. K3’s existence shows that Moonshot can announce a model of this size, but it does not establish what hardware trained it, how much compute it used or how export restrictions affected development. Because K3 is a sparse mixture-of-experts model, its 2.8 trillion total parameters cannot be treated as its per-token compute requirement.
“Our most capable model to date, with 2.8 trillion parameters.”
— Moonshot AI launch materials
Build a Large Language Model (From Scratch)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
License and Efficiency Questions Persist
K3 is being described as an open-weight model, but its weights were not available at launch. Moonshot says they will arrive by July 27. Until then, outside researchers cannot inspect the files, reproduce deployment claims or determine whether the release meets common expectations for open-weight access.
The license and technical report remain unpublished, and Moonshot has not disclosed the model’s active parameter count. The one-million-token context figure is a stated maximum, not proof that every service tier will expose that capacity. Only the Max reasoning setting was available at launch, leaving the behavior and cost of other reasoning levels unresolved.
Optimizing Small Language Models for Production Systems: Designing, Training, Quantizing, and Deploying Lightweight Transformer Models with Python, LoRA, and Modern Compression Techniques
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
July 27 Tests Weight Promise
Attention now turns to Moonshot’s July 27 deadline for releasing K3’s weights. Researchers will examine the license, model files and hardware requirements to determine how freely the system can be used, modified and deployed.
Further independent testing should show whether K3’s early benchmark position holds across coding, reasoning, multimodal and long-context workloads. Publication of the technical report and active parameter count would also clarify the model’s efficiency and the resources needed to run it.
Autel MaxiSYS Ultra S2 AI Scanner, 2026 Top Intelligent Scan Tool V2.0 of MS919 S2/ MS909 S2 MSUltra, 6in1 VCMI2, Topology 3.0 Multi-Point DVI, EV Test, Motor Truspeed, 48+ Service, ECU Pr0gram, OS13
🔥🔥🔥【2026 Autel Ultra S2 AI Scanner with 2 Years Update, V2.0 of MS919 S2/ MS909 S2】Autel unveil the…
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 went live on July 16, 2026, through the Kimi app, Playground and API. Moonshot says downloadable weights will follow by July 27.
How much does Kimi K3 cost?
The API price is $3 per million input tokens and $15 per million output tokens, with a reported cached-input rate of $0.30. Those main rates match Claude Sonnet 5’s stated list price.
Is Kimi K3 already open-weight?
No weights were available at launch. Moonshot has promised a July 27 release, but the license had not been published, so permitted uses remain unknown.
How close is K3 to leading AI models?
Artificial Analysis scored K3 at 57.1 on Intelligence Index v4.1, placing it 2.8 points behind the cited frontier. That is a strong independent result, though no single benchmark establishes overall superiority.
Can K3 run on a personal workstation?
The announced 2.8-trillion-parameter scale points to demanding infrastructure, even with sparse expert routing. Without the weights, active parameter count and deployment specifications, practical workstation requirements cannot yet be confirmed.
Source: Thorsten Meyer AI
