Moonshot’s Kimi K3 aims for Opus 4.8 parity

Anthropic and OpenAI spent most of 2026 trading the frontier crown, but Moonshot AI is crashing the party.

Reports emerged on July 16 that the Chinese AI lab is preparing to launch Kimi K3, an open-weights model packing between 2 trillion and 3 trillion parameters. That makes it the largest open-weight model out of China to date.

Taking on closed-source giants

According to the Financial Times, Kimi K3 is expected to match or even beat Anthropic's Opus 4.8. Other reports indicate the model scored a 57 on AA's Intelligence Index, sitting right behind Claude Fable 5 at 60 and GPT-5.6 Sol at 59, representing a massive double-digit jump over K2.6.

K3 features a 1M context window and reportedly outperforms both Fable 5 and Sol in web research, spreadsheet work, frontend design, and long-form coding.

Moonshot also showcased a demo where K3 ran solo for 48 hours to design and verify a miniature chip. In simulation, the chip ran a mini version of K3 at 8,700 tokens per second.

API pricing is set at $3 per million input tokens and $15 per million output tokens, matching Claude 5 Sonnet. Moonshot plans to release the model weights by July 27.

Alongside the launch, Moonshot is reportedly raising fresh funding at a $31.5 billion valuation, just two months after securing $2 billion at a $20 billion valuation in May.

Why it matters

For enterprise builders and developers, K3 represents a viable escape hatch from expensive frontier lock-in. If your workflow leans heavily on complex coding, data analysis, or web research, K3 matches flagship closed models at a fraction of the cost. Once weights drop on July 27, teams worried about data privacy can bring those weights in-house instead of piping sensitive data into proprietary APIs. The catch? Running a 2-to-3-trillion parameter monster on-prem is going to require serious hardware, keeping smaller shops on managed endpoints.

For the broader industry, this narrows the frontier gap dramatically. The conventional wisdom that open-weights or non-US labs trail the frontier by 6 to 12 months is starting to look outdated; K3 pulls within striking distance inside a single product cycle. We think aggressive open-weights releases like this will force OpenAI and Anthropic to defend their premium pricing. Still, Moonshot has to prove it can support a $31.5 billion valuation while eating the eye-watering compute bills required to train and host models of this size.