Moonshot AI released Kimi K3, a 2.8 trillion parameter model that becomes the world's first open 3T-class frontier intelligence system.

The model features native vision capabilities and a 1 million token context window. While trailing proprietary models Claude Fable 5 and GPT 5.6 Sol in overall performance, Kimi K3 demonstrated frontier-level results across the company's evaluation suite.

Architecture and efficiency gains

Kimi K3 builds on Kimi Delta Attention (KDA) and Attention Residuals (AttnRes), two architectural updates designed to improve information flow across sequence length and model depth.

The model uses Mixture of Experts (MoE) sparsity, activating 16 out of 896 experts when paired with a Stable LatentMoE framework. These structural changes yield approximately 2.5× improvement in scaling efficiency compared to Kimi K2.

For nine of the past twelve months, Kimi models have set the upper bound of open-model sizes, according to the company.

Coding and research capabilities

In kernel optimization tests, Kimi K3 performed competitively with Claude Fable 5 and substantially outperformed Opus 4.8, GPT 5.6 Sol, and GPT 5.5 across GPU tasks spanning NVIDIA Hopper GPUs.

The model built MiniTriton, a compact Triton-like compiler with its own tile-level IR layer over MLIR, optimization passes, and PTX code-generation pipeline. MiniTriton delivers performance on par with or better than Triton and torch.compile on supported benchmarks.

In one research case, Kimi K3 completed computational astrophysics work in two hours that would typically require one to two weeks for an experienced researcher. The model reviewed 20+ papers, implemented numerical pipelines, and generated 3,000+ lines of Python code.

Availability and release timeline

Kimi K3 launches today on Kimi.com, Kimi Work, Kimi Code, and the Kimi API. The model uses maximum thinking effort by default, with low- and high-effort modes coming in subsequent updates.

Full model weights will be released by July 27, 2026. The company is working with inference partners and open-source maintainers to ensure reliable ecosystem rollout.