Kimi K3 and the inversion of the open-weight price curve — the frontier open model is now the expensive one

The open-weight story has been 'nearly frontier capability at a fraction of the cost.' Kimi K3 breaks the pattern by charging a premium for a capability step. When the cheap-and-good-enough model gets a premium sibling, the open ecosystem has stopped being a discount and started being a market.

Kimi K3 pushes the open-weight ceiling higher at roughly three to four times the cost of the prior K2.6, which remains the practical pick for everyday agent work. For once the frontier open model is the expensive one, not the cheap one — and that inversion is the tell.

Tiers are what a market looks like

The open ecosystem is developing a premium flagship and a cheaper workhorse, exactly as the closed labs did. Buyers now choose within a line, not just between vendors. A lab charging a premium for the top of its range is a lab that believes there is a market willing to pay for capability rather than only for cheap-and-good-enough — which is a maturation, not a betrayal, of the open model.

The competitive backdrop is a leaderboard the West largely does not sit on. Four of the five leading open-weight models come from Chinese labs, with Qwen alone accounting for over 40% of new Hugging Face variants. Open source no longer means second best, and the best of it has a distinct centre of gravity.

The eventually-open hedge

K3's weights trailing its launch is a governance tell: ship the API first for revenue, release the weights after to honour the open commitment. That hybrid posture is becoming standard, and it blurs the line between 'open' and 'eventually open' in ways the EU's training-data-summary rule will eventually have to test.

The discount era of open weights is ending. What replaces it is a market — tiered, priced, and, at the top, not cheap.

Tech-Insider — Best open source LLM 2026: DeepSeek, Kimi, Qwen ranked → · Hugging Face — Best open-source LLM models in 2026 →