{"slug": "kimi-k3-has-2-8", "title": "Kimi K3 Has 2.8", "summary": "Moonshot AI released Kimi K3, a 2.8 trillion parameter open-weight model on July 16, 2026, with a 1 million token context window, achieving #1 on LMArena's Frontend Code Arena and outperforming Claude Fable 5. Despite its size, K3 runs inference at the computational cost of a 50 billion parameter model by using only 1.8% of its parameters per token, a design choice detailed in Moonshot's technical report.", "body_md": "Member-only story\n\n# Kimi K3 Has 2.8 Trillion Parameters and Uses 1.8% of Them Per Token. Here Is the Mathematics of Why That Is Genius.\n\nTwo weeks ago the largest open-weight model in existence had one trillion parameters.\n\nThen Moonshot AI shipped Kimi K3.\n\n2.8 trillion parameters. Released July 16, 2026. Open weights. 1 million token context window.\n\n#1 on LMArena’s Frontend Code Arena, beating Claude Fable 5 on the benchmark developers actually care about.\n\nAnd it runs inference with the computational cost of a 50 billion parameter model.\n\nThat last number is the one worth understanding. Not because it is impressive — though it is — but because the mathematics behind it is a specific set of decisions, each one solving a specific problem, and each one documented in Moonshot’s technical report.\n\nThis is not “China’s ChatGPT moment” coverage. This is the four engineering decisions that made K3 possible, derived step by step.\n\n## The Problem With Being Large\n\nScale works. Every lab agrees on this. More parameters, more capability, smoother loss curves, better benchmarks.\n\nThe problem: a 2.8 trillion parameter dense model is not deployable. It is not even runnable.\n\nA dense transformer with 2.8T parameters in float16: 2.8 × 1⁰¹² × 2 bytes = 5.6 terabytes.", "url": "https://wpnews.pro/news/kimi-k3-has-2-8", "canonical_source": "https://pub.towardsai.net/kimi-k3-has-2-8-fe86b6739bd0?source=rss----98111c9905da---4", "published_at": "2026-07-23 03:43:14+00:00", "updated_at": "2026-07-23 03:57:16.741564+00:00", "lang": "en", "topics": ["large-language-models", "ai-research", "ai-products"], "entities": ["Moonshot AI", "Kimi K3", "Claude Fable 5", "LMArena"], "alternates": {"html": "https://wpnews.pro/news/kimi-k3-has-2-8", "markdown": "https://wpnews.pro/news/kimi-k3-has-2-8.md", "text": "https://wpnews.pro/news/kimi-k3-has-2-8.txt", "jsonld": "https://wpnews.pro/news/kimi-k3-has-2-8.jsonld"}}