Realtime AI News
China's Moonshot AI Open-Sources Kimi K3: A 2.8 Trillion Parameter Model That Rivals Frontier Systems
Moonshot AI has open-sourced Kimi K3, China's largest open-weight AI model with 2.8 trillion parameters, a 1-million-token context window, and native vision capabilities. The model scores 57 on the Artificial Analysis Intelligence Index, surpassing Claude Opus 4.8's 56, and leads in benchmarks including frontend coding and browsing agents.
Moonshot AI has open-sourced Kimi K3, China's largest open-weight AI model ever released, featuring 2.8 trillion parameters, a 1-million-token context window, and native multimodal vision capabilities. The move comes as Anthropic tightens model access and OpenAI builds higher walls — Moonshot chose the opposite direction: giving away its flagship model to the world.
On independent benchmarks, Kimi K3 demonstrated frontier-level performance. It scored 57 on the Artificial Analysis Intelligence Index, surpassing Claude Opus 4.8 (56), though still trailing Claude Fable 5 and GPT-5.6 Sol. The model ranked first on AutomationBench (SaaS agent workflows) at 53%, topped Arena's Frontend Code leaderboard ahead of Fable 5, and posted the best public BrowseComp score at 91.2%.
Several architectural innovations power these results. Kimi Delta Attention enables up to 6.3x faster decoding in million-token contexts, while Attention Residuals deliver roughly 25% higher training efficiency at less than 2% additional parameters. These achievements are notable given China's hardware and compute constraints. Bank of America analysts noted that K3 proves "pre-training scaling, paired with architectural innovation, can still deliver step-change gains for flagship Chinese models."
On pricing, Kimi K3 charges $3 per million input tokens and $15 per million output tokens — matching Claude Sonnet 5's pricing exactly. This signals the end of the era when "open-weight" meant "the cheap tier." Moonshot is charging premium prices because the measured intelligence justifies it.
The market reacted immediately. Zhipu AI fell 28.4% and MiniMax dropped 15.6% the following day, reflecting a market repricing of what open-source models can achieve.
Kimi K3's open-source decision carries strategic weight. At a time when leading AI labs are increasingly closing their models, Moonshot's choice to fully open its flagship — challenging top US systems on both benchmarks and price — provides the global developer community with a genuine frontier alternative. The open weights are scheduled for release on July 27.
Why it matters
Kimi K3's open-source release shatters the narrative that frontier models must be closed, proving Chinese teams can reach world-class levels under compute constraints while offering the global AI community a genuine open alternative.
Nearby Updates
All07/21, 00:59
Google Reportedly Developing 'Frozen v2' Chip That Hardcodes Gemini Architecture into Silicon
Google is developing a next-generation AI chip called Frozen v2 that hardcodes the Gemini architecture directly into silicon, achieving up to 10x efficiency improvement per unit of power. The report, originally from The Information, was widely covered by CNBC and Bloomberg, sending Alphabet shares higher.
07/21, 00:06
Kimi K3 smashes open-source ceiling, global tech stocks lose $470B in 3 days
Moonshot AI's Kimi K3 open-weight model triggered a $470 billion tech stock selloff over three days as its near-frontier performance challenged the dominance of proprietary AI models and reshaped market expectations.
07/21, 00:05
Hugging Face forced to use open-source GLM 5.2 for defense after commercial AI models refuse attack data
After an AI agent attack, Hugging Face found that commercial frontier models refused to process real attack data due to safety guardrails, forcing it to switch to Z.ai's open-weight GLM 5.2 for log analysis and defense.
07/21, 01:48
OpenAI Executive Warns China Giving Away High-Quality Models For Free Threatens For-Profit AI Companies
A Futurism report reveals an OpenAI executive publicly expressed concern that China's strategy of releasing high-quality AI models for free is making it difficult for for-profit companies to compete. The comments have reignited debate over open-source versus proprietary AI business models.