{"slug": "moonshot-ai-releases-kimi-k3-a-2-8-trillion-parameter-open-weight-model-rivaling", "title": "Moonshot AI Releases Kimi K3, a 2.8-Trillion-Parameter Open-Weight Model Rivaling Top U.S. Systems", "summary": "Chinese AI startup Moonshot AI released Kimi K3 on July 16, a 2.8-trillion-parameter sparse mixture-of-experts model that is the largest open-weight AI system ever built. The model scores near the top of independent leaderboards, trailing only Anthropic's Claude Fable 5, and signals that Chinese AI labs are closing the gap with U.S. frontier developers despite semiconductor export controls.", "body_md": "# Moonshot AI Releases Kimi K3, a 2.8-Trillion-Parameter Open-Weight Model Rivaling Top U.S. Systems\n\n- Kimi K3 contains 2.8 trillion total parameters in a sparse MoE architecture with a 1-million-token context window, making it the largest open-weight model ever released\n[[1]](https://venturebeat.com/technology/chinas-moonshot-ai-releases-kimi-k3-the-largest-open-source-model-ever-rivaling-top-u-s-systems) - Artificial Analysis reported K3 achieved an overall Elo of 1,547, trailing only Claude Fable 5 and leading Arena.ai's frontend code benchmark\n[[2]](https://simonwillison.net/2026/Jul/16/kimi-k3/) - API pricing is set at $3 per million input tokens and $15 per million output tokens — roughly half the cost of Anthropic's Opus 4.8\n[[3]](https://felloai.com/kimi-k3/) - Moonshot AI raised $2 billion at a $20 billion valuation in May and is now seeking $30 billion in fresh talks\n[[4]](https://techcrunch.com/2026/05/07/chinas-moonshot-ai-raises-2b-at-20b-valuation-as-demand-for-open-source-ai-skyrockets/) - Full open weights are expected to be published by July 27, with rival AI stocks Z.ai and MiniMax falling 27% and 16% respectively after the announcement\n[[5]](https://cryptobriefing.com/moonshot-kimi-k3-ai-semiconductor-stocks-decline/)\n\nChinese AI startup Moonshot AI released Kimi K3 on July 16, a 2.8-trillion-parameter sparse mixture-of-experts model that the company says is the largest open-weight AI system ever built. Two variants — K3 Max for chat and agent tasks, and K3 Swarm Max for parallel processing — went live on the Kimi app, web playground, and API with no formal keynote or press conference [1].\n\nThe model scores near the top of independent leaderboards. Artificial Analysis reported an overall Elo rating of 1,547, placing K3 behind only Anthropic's Claude Fable 5 and ahead of OpenAI's GPT-5.5. It leads Arena.ai's frontend code benchmark, surpassing Claude Fable 5. Moonshot's own benchmarks claim K3 \"mostly beats Claude Opus 4.8 max and GPT-5.5 high,\" though it trails Claude Fable 5 and GPT-5.6 Sol [2].\n\nThe release is the latest and most dramatic signal that Chinese AI labs are closing the gap with U.S. frontier developers despite three years of escalating semiconductor export controls. Moonshot, backed by Alibaba, Tencent, and Meituan, raised $2 billion at a $20 billion valuation in May and is now in talks for a round that would value the company at $30 billion [4].\n\n## The Model\n\nKimi K3 is a sparse mixture-of-experts model, meaning only a fraction of its 2.8 trillion parameters activate on any given request, reducing inference costs compared to a dense model of equivalent size. It is roughly 2.8 times the size of its predecessor, K2.6, and dwarfs Chinese competitors including DeepSeek's V4 Pro (1.6 trillion parameters) and Zhipu AI's GLM 5 series (744 billion parameters) [5].\n\nThe model introduces two architectural innovations: Kimi Delta Attention (KDA) and Attention Residuals (AttnRes), which Moonshot says improve efficiency and reasoning quality. K3 also supports a 1-million-token context window, positioning it for long-horizon coding and agent workloads. Moonshot says K3 uses 21% fewer output tokens than K2.6 on equivalent tasks [2].\n\nAPI pricing is set at $3 per million input tokens and $15 per million output tokens — the highest pricing from any Chinese AI lab, but roughly half the per-task cost of Anthropic's Opus 4.8. The model currently offers only one reasoning effort level ('max'), and independent testers have noted heavy reasoning token consumption — 13,241 tokens for a simple SVG pelican generation, costing about $0.25 per query [3].\n\n## How It Compares\n\nOn Artificial Analysis's composite leaderboard, K3 achieved an Elo of 1,547 — a 732-point jump from Kimi K2.6 and behind only Claude Fable 5. It leads Arena.ai's frontend code benchmark. Moonshot's self-reported results claim parity or superiority over Claude Opus 4.8 and GPT-5.5 on several evaluations, though the model trails both Claude Fable 5 and GPT-5.6 Sol on aggregate measures [2].\n\nIndependent verification remains limited. As Simon Willison noted, no public model card, license file, or downloadable weights were available at launch. Moonshot has promised to release full open weights by July 27, which will allow the broader research community to run its own evaluations [2].\n\nThe rollout was notably informal — described by one observer as 'no keynote, no model card, just a quiet overnight flip of kimi.com' — echoing the low-key release strategy that DeepSeek used to introduce R1 in January 2025 [6].\n\n## Market and Geopolitical Reactions\n\nRival AI stocks fell sharply following the announcement. Chinese AI company Z.ai dropped approximately 27%, while MiniMax fell roughly 16%. Analysts noted that K3's performance-to-cost ratio intensified concerns about pricing power across the AI sector [5].\n\nThe release coincided with Chinese President Xi Jinping's remarks at the 2026 World Artificial Intelligence Conference, where he said AI development should not be a 'solo performance' by one country and pledged support for a 29-nation World AI Cooperation Organization [7].\n\nIn Washington, the timing is awkward. Rep. Young Kim participated in a House Foreign Affairs Committee hearing on July 14 examining export control gaps in the 'AI arms race.' K3's performance underscores a recurring pattern: three years of escalating restrictions on GPUs and lithography equipment have not prevented Chinese labs from reaching or nearing the frontier [8].\n\n## The Company\n\nMoonshot AI, headquartered in Beijing, was founded by Yang Zhilin, a former Google researcher. The company's Kimi chatbot has become one of China's most popular consumer AI products, and its annualized recurring revenue exceeded $200 million in April, driven by paid subscriptions and API usage [4].\n\nThe startup's investor base reads as a who's who of Chinese tech: Alibaba, Tencent, Meituan (which led the May round), HongShan (formerly Sequoia China), ZhenFund, IDG Capital, and 5Y Capital. Total funding stands at approximately $3.77 billion across four rounds. Bloomberg reported in June that Moonshot is seeking a $30 billion valuation in new funding talks — a nearly eight-fold increase from its $4 billion valuation in late 2024 [4].\n\n## What's Next\n\nThe July 27 open-weight release will be the critical inflection point. If independent evaluations confirm Moonshot's benchmark claims, K3 will become the most capable freely available model in the world, accessible to any developer or company without API dependency. That prospect has already shifted the conversation around AI sector valuations [5].\n\nFor U.S. policymakers, the model raises uncomfortable questions. Nvidia has begun limited H200 chip shipments to China after the Commerce Department approved roughly 10 Chinese companies for restricted access. Whether tighter controls can meaningfully slow a lab that just built a 2.8-trillion-parameter model remains an open and increasingly urgent question [7].\n\n## Companies mentioned\n\n## Further sources\n\n[[1] VentureBeat: China's Moonshot AI releases Kimi K3, the largest open-source mode… ↗](https://venturebeat.com/technology/chinas-moonshot-ai-releases-kimi-k3-the-largest-open-source-model-ever-rivaling-top-u-s-systems)\n\n[[2] Simon Willison: Kimi K3, and what we can still learn from the pelican benchmark ↗](https://simonwillison.net/2026/Jul/16/kimi-k3/)\n\n[[3] Fello AI: Kimi K3: Moonshot's 2.8T Open-Weight Model Explained ↗](https://felloai.com/kimi-k3/)\n\n[[4] TechCrunch: China's Moonshot AI raises $2B at $20B valuation ↗](https://techcrunch.com/2026/05/07/chinas-moonshot-ai-raises-2b-at-20b-valuation-as-demand-for-open-source-ai-skyrockets/)\n\n[[5] CryptoBriefing: Moonshot AI's Kimi K3 sends rival stocks tumbling as China's AI… ↗](https://cryptobriefing.com/moonshot-kimi-k3-ai-semiconductor-stocks-decline/)\n\n[[6] Medium / Krishna Kumar: Kimi K3 Max Is Here: Moonshot's 2.8-Trillion-Parameter … ↗](https://xalgord.medium.com/kimi-k3-max-is-here-moonshots-2-8-trillion-parameter-9084f4e4e633)+2 more\n\nThe stories that matter, in one email. Free — unsubscribe anytime.", "url": "https://wpnews.pro/news/moonshot-ai-releases-kimi-k3-a-2-8-trillion-parameter-open-weight-model-rivaling", "canonical_source": "https://mlq.ai/news/moonshot-ai-releases-kimi-k3-a-28-trillion-parameter-open-weight-model-rivaling-top-us-systems/", "published_at": "2026-07-17 12:51:12.971119+00:00", "updated_at": "2026-07-17 12:51:14.744588+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-products", "ai-startups", "ai-policy"], "entities": ["Moonshot AI", "Kimi K3", "Anthropic", "OpenAI", "Alibaba", "Tencent", "Meituan", "DeepSeek"], "alternates": {"html": "https://wpnews.pro/news/moonshot-ai-releases-kimi-k3-a-2-8-trillion-parameter-open-weight-model-rivaling", "markdown": "https://wpnews.pro/news/moonshot-ai-releases-kimi-k3-a-2-8-trillion-parameter-open-weight-model-rivaling.md", "text": "https://wpnews.pro/news/moonshot-ai-releases-kimi-k3-a-2-8-trillion-parameter-open-weight-model-rivaling.txt", "jsonld": "https://wpnews.pro/news/moonshot-ai-releases-kimi-k3-a-2-8-trillion-parameter-open-weight-model-rivaling.jsonld"}}