Moonshot AI's open-source K3 matches top US models, pressuring pricing
A Beijing-based startup backed by Alibaba has released an open-source AI model that benchmarks show rivals the best proprietary systems from Anthropic and OpenAI, threatening to erode pricing power across the industry.
Moonshot AI on Thursday released Kimi K3, a 2.8-trillion-parameter model that the company claims is the largest open-source AI model ever built. Benchmark results from analytics firm Artificial Analysis show the model trading blows with Claude Fable 5 Max and GPT-5.6 Sol Max across real-world task evaluations, long-horizon knowledge work, and complex information retrieval. The full model weights are set to be released on July 27.
The release directly challenges the pricing structure underpinning Western AI labs. Moonshot is offering K3 through its API at $3 per million input tokens and $15 per million output tokens — roughly in line with mid-tier Western offerings, but at performance levels the company says approach the top of the market. Cached input tokens cost just $0.30 per million. The model is also compatible with the OpenAI SDK, meaning developers can switch with minimal friction.
For investors in U.S. AI companies, the implications are stark. The gap between open-source and proprietary models — long cited as a moat justifying premium valuations for companies like Anthropic and OpenAI — appears to have collapsed. On GDPval-AA v2, a benchmark measuring tasks across 44 occupations and 9 industries, K3 scored 1,687, placing it ahead of Claude Opus 4.8 at 1,600. On Artificial Analysis's private agentic benchmark, K3 ranked second overall.
A strategic pivot born of necessity
K3 is as much a business survival story as a technical achievement. Moonshot AI, founded in 2023 by Yang Zhilin, had raised roughly $1.5 billion and seen its valuation climb from $2.5 billion to $4.3 billion, reportedly seeking a new round at $5 billion. Then DeepSeek's low-cost R1 model upended the Chinese AI market in January 2025. Moonshot's Kimi platform slid from third to seventh in monthly active users in China.
The company's pivot to open-source — beginning with Kimi K2 in July 2025 — was an effort to reclaim relevance. K3, with its 2.8 trillion parameters roughly 75 percent larger than DeepSeek's V4 Pro, represents the culmination of that strategy.
Geopolitical calculus
The decision to open-source the weights also serves Beijing's broader strategic interests. Releasing open-source models allows Chinese companies to showcase capabilities and expand global developer communities, a strategy that helps counter U.S. efforts to restrict China's tech progress. Moonshot's own timeline charts position K3 far above the largest open-source releases from DeepSeek, Xiaomi, and Alibaba.
Beyond benchmarks, Moonshot demonstrated K3 autonomously designing a functional chip over 48 hours of continuous operation, and reproducing a complex astrophysics calculation in two hours that would typically take a senior researcher one to two weeks. These point toward autonomous agent capabilities that could reshape enterprise workflows — the very market Western labs are targeting for revenue growth.
A promotional rebate offering up to 30 percent back on API credits of $1,000 or more runs through August 12, signalling an aggressive push for developer adoption in the critical weeks after release.
