Crimson AI NewsA CrimsonLingua Network service
EN ع
← Back to news
Kimi (Moonshot)

Moonshot AI Unveils Kimi K3: World's First Open 3T-Class Model with 2.8T Parameters

AI By Crimson AI Kimi Blog 17 August 2026 · 12:32 12 views
Share: X Telegram

Moonshot AI releases Kimi K3, a 2.8T-parameter open model with native vision and a 1M-token context, claiming frontier-level performance in coding and reasoning, with weights to be released by July 27, 2026.

Moonshot AI Unveils Kimi K3: World's First Open 3T-Class Model with 2.8T Parameters

Key points

Moonshot AI has introduced Kimi K3, its most capable model to date, positioning it as the world's first open 3T-class model. With 2.8 trillion parameters, Kimi K3 is built on novel architectural components—Kimi Delta Attention (KDA) and Attention Residuals (AttnRes)—and features native vision capabilities and a 1-million-token context window. The company states that while overall performance still trails proprietary leaders like Claude Fable 5 and GPT 5.6 Sol, Kimi K3 consistently outperforms other tested open models across their evaluation suite.

Kimi K3 is available immediately on Kimi.com, Kimi Work, Kimi Code, and the Kimi API. At launch, it defaults to 'max thinking effort,' with low- and high-effort modes planned for later updates. The full model weights are scheduled for release by July 27, 2026, with a technical report detailing architecture, training, and evaluations to follow. Moonshot is collaborating with inference partners and open-source maintainers to ensure a smooth ecosystem rollout.

The model's architecture introduces KDA and AttnRes to improve information flow across sequence length and depth, combined with a Mixture of Experts (MoE) design that activates 16 of 896 experts via a Stable LatentMoE framework. These changes yield an approximate 2.5× improvement in scaling efficiency over Kimi K2, enabling more effective conversion of compute into intelligence.

In benchmarks, Kimi K3 demonstrated strong long-horizon coding performance, including autonomous GPU kernel optimization, where it competed with Fable 5 and outperformed Opus 4.8, GPT 5.6 Sol, and GPT 5.5. It also built MiniTriton, a Triton-like compiler from scratch, achieving performance on par with or better than Triton and torch.compile on roofline benchmarks, and sustaining end-to-end nanoGPT training.

Beyond coding, Kimi K3 excels in vision-language tasks, 3D reasoning, and scientific research. It autonomously designed a chip for a nano model in 48 hours, and reproduced I-Love-Q universal relations in astrophysics in about two hours—a task typically requiring one to two weeks. In Kimi Work, new Widgets and Dashboard features enable interactive visualizations and persistent, personalized views.

ModelParametersContext WindowAvailability
Kimi K32.8T1M tokensNow (weights by Jul 27, 2026)
Source
Kimi (Moonshot) · Kimi Blog
Related news
Research paper
Kimi (Moonshot) 17 Aug 2026

Moonshot AI Unveils Kimi K2.5: Open-Source Visual Agentic Intelligence with Swarm Capabilities

Moonshot AI introduces Kimi K2.5, a native multimodal open-source model with advanced coding and vision, featuring a self-directed...

22
Research paper
Kimi (Moonshot) 17 Aug 2026

Moonshot AI Unveils WorldVQA Benchmark to Test Visual World Knowledge in Multimodal LLMs

Moonshot AI releases WorldVQA, a benchmark with 3,500 image-question pairs designed to measure factual visual knowledge in multimo...

18
Research paper
Kimi (Moonshot) 17 Aug 2026

Kimi Launches Agent Swarm: 100 AI Agents Self-Organize to Tackle Complex Tasks

Kimi (Moonshot) unveils Agent Swarm, a research preview that lets K2.5 deploy up to 100 parallel sub-agents that self-organize int...

15