Open weights at frontier scale — Kimi K3 and the closing gap

A 2.8-trillion-parameter open-weight model with native vision is not a fallback. It is the open ecosystem matching the largest closed labs on scale and shipping the weights anyway.

Moonshot shipped Kimi K3, a 2.8-trillion-parameter open-weight mixture-of-experts model with native vision and a 1-million-token context. The parameter count is a statement of intent: the open labs are no longer trailing on scale, they are matching the largest closed systems — and the mixture-of-experts design keeps a model this big deployable by activating only a fraction per token.

The ecosystem compounds

Shipping vision and long-context in the base release means every fine-tune and product built on K3 inherits them for free. That is the compounding advantage that turned the leading open models into default foundations rather than alternatives — capability plus open weights seeds a whole downstream ecosystem.

And it is a full product line now, not a hero model. Qwen 3.7 Flash, GLM-5.2, DeepSeek V4 Pro, and MiniMax M3 fill flagship, flash, and coding tiers at four-to-ten-times-lower cost. The open ecosystem mirrors how the closed labs segment — a point on the capability-cost curve for every workload.

The pressure that drives the price war

When open weights sit single-digit percentage points behind premium models at a fraction of the cost, the closed labs must justify their price with a capability lead or a matching cut. A frontier-scale open release like K3 sharpens the question every buyer now asks: what exactly is the closed premium still buying?

LLM-Stats — AI updates today (August 2026) → · Taskade — 10 best open-source LLMs in July 2026 →