TL;DR
Moonshot AI released Kimi K3 on July 16 with a 2.8-trillion-parameter architecture, near-frontier independent benchmark results and API prices matching Claude Sonnet 5’s standard rate. The launch could weaken the view of Chinese models as discounted alternatives, but the promised weights, licence and technical report are not yet public.
Moonshot AI released Kimi K3 on July 16, introducing a 2.8-trillion-parameter model that independent testing placed close to leading Western systems while charging $3 per million input tokens and $15 per million output tokens. The combination of stronger performance and higher pricing matters because it suggests Moonshot is positioning K3 as a direct capability competitor, not primarily as a discounted Chinese alternative.
Kimi K3 is available through the Kimi app, Playground and API, according to Moonshot’s launch materials. The company describes it as its most capable model, with 2.8 trillion total parameters, a sparse mixture-of-experts architecture that routes 16 of 896 experts per token, and a maximum context window of 1,048,576 tokens. It accepts text, image and video inputs, although only the Max reasoning setting was available at launch.
Artificial Analysis’ Intelligence Index v4.1 gave the announced open-weight configuration a score of 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. The testing group also recorded a 732-point long-horizon Elo increase from Kimi K2.6 to K3, placing the new model at 1547. These are historical benchmark results and do not guarantee performance across production workloads.
The pricing marks an equally important change. Thorsten Meyer AI calculated that K3 costs about five times as much as its predecessor and matches Claude Sonnet 5’s standard $3 and $15 list prices. Sonnet 5’s temporary introductory rate of $2 for input and $10 for output, available through August 31, leaves K3 about 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.
Chinese AI Drops Its Discount
Chinese AI providers have often competed through a mix of lower API prices, downloadable weights and improving capability. K3 changes that equation by asking customers to pay a Western mid-tier price before its weights and licence can be inspected. Thorsten Meyer AI interprets the decision as a sign that Moonshot believes K3 can compete on performance without a large price concession.
If customers accept that price, Western developers may face stronger competition on model quality rather than cost alone. K3’s scale also complicates claims that hardware export restrictions have kept Chinese laboratories far from the frontier. Still, model size does not reveal training efficiency: K3 is sparse, its active parameter count is undisclosed, and the launch materials do not establish which hardware or computing resources were used.
Vision-Language Models in Production: Architecting Multimodal LLM Applications: From Vision-Language API to Self-Hosted Model (Production AI Engineering Series)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
K3 Accelerates Moonshot’s Climb
K3 follows Moonshot’s K2 family, whose approximate API rate was $0.60 per million input tokens and $3 per million output tokens, according to the supplied analysis. Its announced 2.8-trillion-parameter scale is nearly three times K2’s total size and exceeds the cited 1.6-trillion-parameter DeepSeek V4-Pro. Because mixture-of-experts models activate only part of their architecture for each token, total parameters cannot be treated as a direct measure of computing cost or speed.
“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.
Licence and Efficiency Data Missing
K3 cannot yet be independently described as a released open-weight model because the weights are not available. Moonshot has promised them by July 27, but the licence and technical report remain unpublished. Until those materials appear, developers cannot verify whether commercial use, modification and redistribution will be permitted on terms comparable with permissive open licences.
Moonshot has also not disclosed the active parameter count, detailed training resources or broad production evaluations. The one-million-token figure is a maximum, and some service tiers may impose lower limits; the supplied material says Moderato is capped at 256,000 tokens. It is also unclear whether K3’s benchmark gains will carry over to reliability, latency and operating cost in routine deployments.
Language Translator Device, Voice Ai Translator Device T21 Wi-Fi Available, Two-Way 138 Languages 2026 Instant Translator Offline Recording Photo Translation Device for Business Travel Study
【Ai Language Translator Device for All Languages Supporting 138 Languages】Our spanish translator device T21 adopts the latest technology…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
July 27 Becomes the Test
Attention will turn to Moonshot’s promised July 27 weight release. The licence, model files and technical report will determine whether K3’s open-weight description is supported and whether outside researchers can reproduce its results. Developers will also watch for additional reasoning settings, independent safety tests and real-world pricing comparisons as usage expands.
DULIWO Scribing Tool Kit for Gunpla – 7-Blade Model Scriber Chisel Set (0.1–2.0mm), Pin Vise Hand Drill with 10 Bits, Tweezers & Brush Included for Gundam HG/RG/MG, Resin Kits, Panel Line Engraving
Model Kit Tools: Includes metal scribe tool (1 handle + 7 blades), small hand drill set (10 bits…
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, announced with 2.8 trillion total parameters and support for text, image and video inputs. It is currently accessible through Moonshot’s app, Playground and API.
How close is Kimi K3 to leading Western models?
Artificial Analysis scored K3 at 57.1 on its Intelligence Index v4.1, 2.8 points below the highest comparison cited in the source material. That places it close on this benchmark, but one index cannot establish equal performance across all tasks.
How much does Kimi K3 cost?
The API price is $3 per million input tokens and $15 per million output tokens, with a cited cached-input rate of $0.30. Usage patterns, context length and reasoning settings can change the final cost.
Is Kimi K3 open source now?
No model weights or licence had been published at the time of the supplied report. Moonshot says weights are due by July 27, so claims about open access remain unconfirmed pending the release terms.
Source: Thorsten Meyer AI
