Price per token flattened in June after climbing nearly 20% in May. Vercel’s AI Gateway Production Index shows open-weight models ran 29% of all gateway tokens — up from 11% in April — on under 4% of total spend.
The numbers from June 2026:
- Token volume grew 29%; spend grew 27% — nearly identical rates after May’s 2x divergence
- Open-weight share: 11% in April → 29% in June, a 2.6x jump in two months
- DeepSeek reached 22.6% of token volume, third place among all models routed through the gateway
- Cost share: open-weight models carried nearly a third of traffic on under 4% of spend
Developers are routing production volume to models that cost a fraction of frontier alternatives. DeepSeek V4 Flash became the first open-weight model that teams drop directly into agentic pipelines as a plausible Anthropic substitute, according to OpenRouter’s June analysis. GLM-5.2 is also climbing.
Open-weight capability has held a consistent 3-6 month gap behind US frontier labs for over 18 months. That gap is now narrow enough that price dominates the routing decision for a large chunk of production traffic.
NOTE
Vercel’s index reflects production traffic routed through AI Gateway — tens of trillions of tokens monthly between enterprise applications and AI labs. It captures what teams actually deploy, not benchmark preferences.
The price floor held. If open-weight volume keeps climbing at this rate and spend doesn’t follow, the per-token cost curve bends downward — and the routing math for frontier models gets harder to justify on non-critical paths.