Moonshot AI just dropped what might be the largest open-weight AI model to come out of a Chinese lab. Kimi K3, which launched on July 16, packs 2.8 trillion parameters and a 1 million token context window, putting it in direct competition with the likes of Claude Fable 5 and GPT-5.6 Sol.

The open weights release is scheduled for July 27 under a Modified MIT license.

What Kimi K3 actually brings to the table

Kimi K3 introduces something called Kimi Delta Attention, which allows for up to 6.3 times faster decoding compared to standard approaches. A feature called Attention Residuals reportedly improves training efficiency by roughly 25% over its predecessor, K2.6.

Early evaluations place Kimi K3 second in performance metrics when benchmarked against top-tier western models. That said, independent benchmarks beyond provider-reported scores remain largely absent, so those rankings deserve a healthy grain of salt until third-party testing catches up.

On pricing, Moonshot AI is coming in competitive: $3 per million input tokens and $15 per million output tokens. That’s deliberately aligned with western pricing standards rather than undercutting them, suggesting Moonshot AI is positioning Kimi K3 as a quality play rather than a budget alternative.

Why crypto cares about open-weight AI models

Open-weight models are the raw material for decentralized inference networks, fine-tuning marketplaces, and AI agent platforms built on blockchain infrastructure. The Modified MIT license is permissive enough to allow commercial use, which means decentralized protocols can integrate Kimi K3 without licensing complications.

Kimi K3 is designed to excel at agentic tasks and long-context reasoning, two capabilities that are particularly relevant for crypto’s AI agent trend.

The broader competitive landscape

No Chinese AI lab has previously offered an open-weight model at this scale. Previous Kimi models, including K2 and K2.6, established the company’s reputation for strong long-context reasoning capabilities.

The risk, as always with provider-reported benchmarks, is that real-world performance doesn’t match the marketing. Independent evaluations will be critical in determining whether Kimi K3 genuinely competes at the Claude and GPT tier or falls short under rigorous testing.