Blab AI
Moonshot AI Launches 2.8-Trillion-Parameter Kimi K3, Challenging OpenAI and Anthropic
Moonshot AI, a Beijing‑based startup backed by Alibaba and Tencent, unveiled its newest large‑language model, Kimi K3, on July 16 2026. The 2.8‑trillion‑parameter model becomes the first open‑weight model to breach the 3‑trillion‑parameter threshold. It also ships with a one‑million‑token context window and native visual‑understanding capabilities, positioning it as a multimodal competitor in the high‑performance LLM arena.
Kimi K3 relies on a hybrid attention architecture that blends Kimi Delta Attention with Attention Residuals, wrapped around a mixture‑of‑experts backbone. The MoE design activates only a subset of the 2.8 trillion parameters for each task, keeping the model computationally efficient while delivering high performance. Quantization with MXFP4/MXFP8 techniques further reduces inference cost, and the model supports the full 1 M‑token context window.
In benchmarks released by Moonshot AI, Kimi K3 topped OpenAI’s GPT‑5.6 Sol and Anthropic’s Claude models on several front‑end coding and agentic/coding tasks. The company also noted that the model ranked third on the Artificial Analysis Intelligence Index, a broad‑range AI benchmark leaderboard.
Pricing for the Kimi K3 API sits below GPT‑5.6 Sol, with the startup reporting a 40‑50 % discount on comparable usage. The lower price point is aimed at developers who need large‑context, multimodal capabilities without the higher cost of leading commercial models.
The launch has drawn attention from investors and market analysts because it introduces a new high‑performance LLM competitor. Anthropic, which recently closed a $65 billion Series H round at a $965 billion valuation, has set a target of $1.2 trillion by the end of 2026. Current market pricing models factor Kimi K3’s capabilities into Anthropic’s valuation, yielding a 91.5 % probability of reaching the $1.25 trillion mark by December 31 2026. Analysts note that Kimi K3 could inject uncertainty into that projection.
Observers will watch Anthropic’s next moves, including potential funding or partnership announcements that could counterbalance the competitive pressure from Kimi K3. The company’s valuation metrics on the Nasdaq Private Market and any subsequent strategic initiatives will serve as key indicators of how the market responds.
Moonshot AI announced that the full weights for Kimi K3 will be released on July 27 2026, after the hosted model has been available since launch. The public documentation lists technical specifications and pricing tiers but does not yet provide checkpoint files. The release of the full weights is expected to broaden access for researchers and developers who prefer an open‑weight model.
In summary, Kimi K3’s entry into the LLM market introduces a high‑parameter, open‑weight model with competitive pricing and strong benchmark performance. Its impact on the valuation of other leading AI companies, particularly Anthropic, remains to be seen as the market adjusts to the new competitive landscape. The forthcoming release of the full weights and continued adoption metrics will offer further insight into Kimi K3’s role in the evolving AI ecosystem.
Kimi K3 relies on a hybrid attention architecture that blends Kimi Delta Attention with Attention Residuals, wrapped around a mixture‑of‑experts backbone. The MoE design activates only a subset of the 2.8 trillion parameters for each task, keeping the model computationally efficient while delivering high performance. Quantization with MXFP4/MXFP8 techniques further reduces inference cost, and the model supports the full 1 M‑token context window.
In benchmarks released by Moonshot AI, Kimi K3 topped OpenAI’s GPT‑5.6 Sol and Anthropic’s Claude models on several front‑end coding and agentic/coding tasks. The company also noted that the model ranked third on the Artificial Analysis Intelligence Index, a broad‑range AI benchmark leaderboard.
Pricing for the Kimi K3 API sits below GPT‑5.6 Sol, with the startup reporting a 40‑50 % discount on comparable usage. The lower price point is aimed at developers who need large‑context, multimodal capabilities without the higher cost of leading commercial models.
The launch has drawn attention from investors and market analysts because it introduces a new high‑performance LLM competitor. Anthropic, which recently closed a $65 billion Series H round at a $965 billion valuation, has set a target of $1.2 trillion by the end of 2026. Current market pricing models factor Kimi K3’s capabilities into Anthropic’s valuation, yielding a 91.5 % probability of reaching the $1.25 trillion mark by December 31 2026. Analysts note that Kimi K3 could inject uncertainty into that projection.
Observers will watch Anthropic’s next moves, including potential funding or partnership announcements that could counterbalance the competitive pressure from Kimi K3. The company’s valuation metrics on the Nasdaq Private Market and any subsequent strategic initiatives will serve as key indicators of how the market responds.
Moonshot AI announced that the full weights for Kimi K3 will be released on July 27 2026, after the hosted model has been available since launch. The public documentation lists technical specifications and pricing tiers but does not yet provide checkpoint files. The release of the full weights is expected to broaden access for researchers and developers who prefer an open‑weight model.
In summary, Kimi K3’s entry into the LLM market introduces a high‑parameter, open‑weight model with competitive pricing and strong benchmark performance. Its impact on the valuation of other leading AI companies, particularly Anthropic, remains to be seen as the market adjusts to the new competitive landscape. The forthcoming release of the full weights and continued adoption metrics will offer further insight into Kimi K3’s role in the evolving AI ecosystem.
