TL;DR
Moonshot AI released Kimi K3 on July 16 at the same list price as Claude Sonnet 5, departing from the low-cost positioning associated with Chinese AI models. Early independent testing places K3 close to leading systems, but it has not taken the overall lead, and its promised weights, licence and technical report are still pending.
Moonshot AI released Kimi K3 on July 16, pricing the new model at $3 per million input tokens and $15 per million output tokens, matching Claude Sonnet 5’s stated list price. Early independent testing places K3 near leading Western systems, but it has not established an overall performance lead, and the promised model weights and licence are not yet available.
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 for each token. The company has not disclosed the active parameter count, which limits direct comparisons between K3’s total size and the computing demands of competing models.
On the Artificial Analysis Intelligence Index v4.1, K3 scored 57.1, compared with 59.9 for Claude Fable 5 using an Opus 4.8 fallback and 58.9 for GPT-5.6 Sol Max. That places K3 2.8 points behind the highest result in the supplied comparison. It also reached first place on Design Arena and recorded a 732-point Elo increase over K2.6 on Artificial Analysis’s long-horizon tracker, according to Thorsten Meyer AI’s account of the results.
The pricing change may be as consequential as the benchmark scores. Thorsten Meyer AI estimates that K3 costs about five times as much as its predecessor and calls it the most expensive model released by a Chinese laboratory. Although its standard price matches Claude Sonnet 5’s $3/$15 list rate, Sonnet 5 has an introductory $2/$10 rate through August 31, making K3 50% more expensive during that promotion.
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.
Claude-Level Pricing Resets Competition
K3 challenges the idea that Chinese AI laboratories compete mainly by offering lower-cost alternatives to Western models. Moonshot is asking customers to compare K3 on capability at a similar list price, a commercial position that will be tested through real workloads, reliability and customer adoption rather than leaderboards alone.
The model also matters because Moonshot intends to release its weights. If the terms permit broad commercial use, developers could gain a near-frontier open-weight option with text, image and video input. That could pressure proprietary vendors on customization and deployment control, though the licence has not been published and K3 cannot yet be evaluated as an accessible open-weight release.
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.
K3’s Early Benchmark Position
Chinese model developers have recently been associated with competitive performance at lower prices, supported in some cases by downloadable weights. K3 departs from the pricing side of that formula while retaining a promised open-weight release, creating a test of whether buyers will pay Western mid-tier rates for a Chinese model.
Moonshot says K3 supports a maximum context window of 1,048,576 tokens and native text, image and video inputs. Reasoning is always active with an adjustable effort setting, but only the Max setting was available at launch. The one-million-token figure is a maximum specification; actual limits may vary by product tier or service configuration.
“Our most capable model to date, with 2.8 trillion parameters.”
— Moonshot AI launch materials
Licence and Efficiency Questions Persist
Several details needed to judge K3 remain unavailable. Moonshot has not published the model licence, technical report or active parameter count. The weights are promised by July 27, meaning descriptions of K3 as an available open-source or open-weight model are premature as of July 17.
It is also unclear how K3 will perform across independent evaluations, production traffic and lower reasoning settings. Artificial Analysis provides an early external result, but the figures cited are about one day old, while Moonshot’s other benchmark numbers are self-reported. The reported scores show K3 close to the leaders, not outpacing the field overall.
July 27 Release Faces Scrutiny
Attention now shifts to Moonshot’s promised July 27 weights release. Developers will examine the licence, download requirements, hardware demands and whether the published files match the model available through Moonshot’s hosted services.
Independent testers are also expected to check benchmark reproducibility, long-context performance and operating costs. Those findings will determine whether K3 represents a durable competitive shift or an impressive early result with unresolved deployment limits. For now, K3 is a close challenger rather than a confirmed leader.
Key Questions
Is Kimi K3 currently the leading AI model?
No. Early independent testing places K3 near the leaders, but the supplied Artificial Analysis results put it behind the highest-scoring systems. Its first-place Design Arena result shows strength in one evaluation, not an overall market lead.
How much does Kimi K3 cost?
K3 costs $3 per million input tokens and $15 per million output tokens. That matches Claude Sonnet 5’s listed rate, although Sonnet 5’s temporary introductory price is lower through August 31.
Is Kimi K3 open source?
Not at launch. Moonshot has promised model weights by July 27, but the files and licence were unavailable on July 17. Whether K3 qualifies as open source will depend on the actual licence terms.
Why is the 2.8-trillion parameter figure important?
It makes K3 one of the largest open-weight models announced, based on total parameters. Because the architecture activates only a subset of experts for each token and Moonshot has not disclosed active parameters, total size does not reveal its full computing cost.
What evidence would confirm K3’s competitive position?
The strongest evidence would include repeated independent benchmark results, production reliability data and testing of the released weights. Reviewers will also need the technical report and licence to judge performance, efficiency and deployment freedom.
Source: Thorsten Meyer AI
