Moonshot’s Kimi K3 Costs Half as Much as GPT-5.6 Sol — and the Price War Just Went Nuclear
Moonshot AI’s Kimi K3 is the largest open-weight model ever built — 2.8T parameters, half the price of GPT-5.6 Sol, and it just tanked Asian AI stocks.
Moonshot AI shipped Kimi K3 on July 16, 2026, and it immediately reset expectations for what an open-weight model can do — at 2.8 trillion parameters, it is the largest open-weight model released so far, and on independent testing it ranks fourth among all frontier models, trailing only Claude Fable 5 and GPT-5.6 Sol, and ahead of Claude Opus 4.8. That alone would have been a story. But Moonshot paired the spec sheet with a pricing structure that Bank of America analysts described in a Friday note as making K3 “half as expensive as OpenAI’s high-performing GPT-5.6 Sol model” — and suddenly the whole AI industry had a DeepSeek flashback.
Z.ai fell as much as 30% in Hong Kong trading, its worst single-day drop since listing in January; MiniMax Group dropped as much as 16%; Alibaba fell 4%; and Bloomberg’s Asian semiconductor index saw a dip of more than 6%, with Nasdaq 100 futures falling 2%. The question now is whether this is a repeat panic — or whether K3 genuinely rewrites how the frontier AI market works.
The Numbers That Spooked the Market
Kimi K3 costs $3.00 per million input tokens and $15.00 per million output tokens. That is still roughly a third of Claude Fable 5’s list rates of $10 input and $50 output. Against GPT-5.6 Sol at $5 input and $30 output, the token-level math is blunt: K3 is 40% cheaper on input and 50% cheaper on output. Per-task costs tell an even tighter story. K3 costs about $0.94 per weighted Intelligence Index task, compared with $1.04 for GPT-5.6 Sol at max and $2.75 for Fable 5 with fallback.
What makes this structurally different from prior Chinese AI price cuts is that Moonshot is not running a loss-leader play. K3 broke the one rule every Chinese AI lab had followed until now: it is not trying to be the cheap alternative. At $3 input and $15 output per million tokens, Moonshot AI is charging Claude Sonnet money and claiming frontier results to justify it. In other words, Moonshot isn’t discounting its way to adoption — it’s pricing like it belongs at the frontier. Bank of America analyst Alex Liu put it plainly: “K3 raises the capability ceiling for China AI models, shifting the burden of proof to other independent AI labs.”
What K3 Actually Is
Kimi K3 is Moonshot AI’s flagship open Mixture-of-Experts model for long-horizon coding, knowledge work, and reasoning — it has 2.8 trillion total parameters and activates 16 of 896 experts per token. It ships with native vision, a 1-million-token context window, always-on thinking, and an architecture built from the ground up with two new innovations. One of those — Kimi Delta Attention — deserves attention: Moonshot says it enables up to 6.3x faster decoding for million-token contexts, which is what makes 1M-token context practically deployable rather than theoretically possible.
On benchmarks, K3 scored 57.11 on the Artificial Analysis Intelligence Index, placing it fourth overall, behind Claude Fable 5 (59.86), GPT-5.6 Sol max (58.89), and GPT-5.6 Sol xhigh (57.65) — ahead of Claude Opus 4.8, Grok 4.5, and GLM-5.2. On coding specifically, K3 topped the front-end programming leaderboard on Arena AI with a score of 1,679, surpassing Claude Fable 5’s 1,631 points and GPT-5.6 Sol’s 1,618 points. Arena co-founder Anastasios Angelopoulos called it a landmark, according to SiliconAngle: “On Code Arena, Kimi K3 has BEATEN FABLE. This makes Kimi_Moonshot the #1 AI lab in the world on frontend.”
The full model weights are not live yet. Moonshot will publish the full open weights by July 27, 2026. Once they land on Hugging Face under a Modified MIT license, anyone with sufficient hardware — we’re talking enterprise-grade clusters, not a MacBook — can self-host a model that scores within a few points of the best closed systems on Earth. Open weights allow developers to run the model themselves, fine-tune it, and deploy it without paying API fees.
The End of Cheap Chinese AI — and Something Scarier
Historically, Chinese software firms engaged in brutal, margin-killing price wars to acquire market share, driving costs down to fractions of a cent. DeepSeek V4 Pro still costs around $0.04 per task; MiniMax M3 around $0.12. K3 at $0.94 per task is a different creature entirely. Commentators read this as the end of “super-cheap Chinese AI”: Moonshot is now pricing on capability rather than undercutting.
That’s the twist Western labs should find uncomfortable. When Chinese models competed purely on price, the implicit response was: “sure, but you get what you pay for.” K3 removes that comfort. At 2.8 trillion parameters it is the largest open-weight model released so far, and on independent testing it ranks fourth among all frontier models. For anyone building on open models, K3 is the first Chinese release that competes with the top U.S. systems on capability rather than just on price. Morgan Stanley analyst Gary Yu framed K3 as the result of steady compound progress: “K3 has received positive feedback globally, signaling an all-round catch-up of Chinese LLMs with U.S. leaders in model size, performance, and pricing.”
The market reaction also hit companies you wouldn’t immediately expect. Leading chipmaker Taiwan Semiconductor Manufacturing Company fell by 7% on Friday, despite reporting a 77% jump in quarterly operating profit. SoftBank — often seen as a proxy for OpenAI — fell by 9%. The logic: if frontier-quality, open-weight models can be built and priced at mid-tier rates by Chinese labs, the enormous compute spend Western infrastructure companies are counting on suddenly looks a lot shakier.
Not everyone’s convinced it’s catastrophic. Patrick Moorhead, CEO and chief analyst at Moor Insights and Strategy, characterized the market’s reaction as “an over-reaction shockingly similar to the DeepSeek panic,” explaining that despite the technology’s advances, “We are far away from super-intelligence.” Founded in 2023, Moonshot raised $2 billion in May at a valuation above $20 billion, with Alibaba and Tencent among its backers. A new round reportedly targeting a $31.5 billion valuation is said to be in progress, according to TechCrunch.
What’s Next
July 27 is the real date to watch. Moonshot will release full open-source weights — meaning businesses can self-host the model on their own infrastructure — on July 27, 2026, alongside a technical report covering architecture, training, and benchmark results. That’s when independent evaluations will either confirm K3’s benchmark claims or complicate them. Chinese AI has earned a global following by being systematically cheaper than the U.S. competition — DoorDash, for example, is already pushing “lower-level work” to Moonshot’s Kimi model, leading to “better quality, cheaper cost,” according to its CTO. K3 suggests the next phase isn’t just cheaper — it’s also better on the benchmarks that developers actually care about. For OpenAI and Anthropic, the burden of proof just shifted. Being closed-source and expensive used to come with a capability premium. That premium is eroding fast.
