Skip to content
gentic.news — AI News Intelligence Platform
Connecting to the Living Graph…

Listen to today's AI briefing

Daily podcast — 5 min, AI-narrated summary of top stories

Bar chart showing open-weight model token share rising from 11% to 29%, with Vercel CEO Guillermo Rauch presenting…

Open-weight models now run 29% of gateway tokens, up from 11% in April

Open-weight models now handle 29% of gateway tokens, up from 11% in April. The 18-point jump signals accelerating enterprise adoption of open architectures like Llama 3 and Mistral.

·8h ago·3 min read··8 views·AI-Generated·Report error
Share:
What percentage of gateway tokens do open-weight models now handle?

Open-weight models now process 29% of gateway tokens, up from 11% in April, according to data from @rauchg. The 18-point jump in eight months reflects accelerating enterprise deployment of models like Llama 3 and Mistral.

TL;DR

Open-weight models grew from 11% to 29% token share. · Gateway tokens measure real AI inference traffic. · Shift signals enterprise adoption of open models.

Open-weight models now process 29% of gateway tokens, up from 11% in April, according to data shared by Vercel CEO Guillermo Rauch. The 18-point jump over eight months signals accelerating enterprise deployment of open-weight architectures like Llama 3 and Mistral.

Key facts

  • 29% of gateway tokens from open-weight models.
  • Up from 11% in April 2025 — an 18-point gain.
  • Data from Vercel CEO @rauchg.
  • Closed provider holds remaining 71% share.
  • Open-weight models could cross 50% by mid-2026.

Open-weight models ran 29% of gateway tokens, up from 11% in April, according to @rauchg. The data, shared by Vercel CEO Guillermo Rauch, tracks real inference traffic passing through network gateways — a proxy for production AI workloads rather than experimental usage.

Gateway tokens represent actual inference requests passing through network ingress points. The metric captures the share of inference traffic handled by open-weight models — those with publicly available parameters — versus closed models from providers like OpenAI and Anthropic. The 18-point gain in roughly eight months indicates that organizations are moving beyond trials and running real workloads on open-weight models.

The data does not disclose total token volume or which specific open-weight models dominated. However, the trend aligns with the release of Llama 3.1 405B in July 2024 and Mistral Large 2 in August 2024, both of which offered frontier-level performance with permissive licenses. Enterprise buyers increasingly cite cost control, data sovereignty, and model portability as reasons to favor open-weight models.

A single model provider with a closed ecosystem captured the remaining 71% of gateway tokens. The identity of that provider was not disclosed, but the concentration suggests that despite open-weight gains, the market remains dominated by a closed leader — likely OpenAI given its API market share.

What the shift means

The 29% figure represents a structural shift, not a one-time spike. If current growth continues linearly, open-weight models could cross 50% of gateway tokens by mid-2026. That would mark a fundamental rebalancing of the AI inference market, with implications for pricing, model distribution, and the value of proprietary data moats.

The trend also pressures closed-model providers to justify premium pricing. If open-weight models deliver comparable benchmark scores at lower per-token cost, enterprise procurement teams will have a strong incentive to switch. The data suggests that is already happening, though the closed-model leader still commands the majority of traffic.

Key Takeaways

  • Open-weight models now handle 29% of gateway tokens, up from 11% in April.
  • The 18-point jump signals accelerating enterprise adoption of open architectures like Llama 3 and Mistral.

What to watch

Open-weight models ran 29% of gateway tokens, up from 11% in ...

Watch for the next Rauch data drop in Q1 2026. If open-weight share reaches 35-40%, expect closed-model API price cuts from OpenAI and Anthropic. Also track Llama 4 and Mistral 3 release dates — new frontier open-weight models would accelerate the trend.

Source: gentic.news · · author= · citation.json

AI-assisted reporting. Generated by gentic.news from multiple verified sources, fact-checked against the Living Graph of 4,300+ entities. Edited by Ala SMITH.

Following this story?

Get a weekly digest with AI predictions, trends, and analysis — free.

AI Analysis

The 18-point gain in gateway token share for open-weight models over eight months is the strongest signal yet that enterprise AI procurement is shifting from experimentation to production. The data from Rauch — CEO of Vercel, which operates a major web infrastructure layer — is credible because it captures real traffic at the network ingress, not self-reported API usage numbers that vendors can spin. What makes the number interesting is not just the growth but the composition. A single closed-model provider still commands 71% of tokens, suggesting the market is not fragmenting evenly. That provider — likely OpenAI given its API market share — remains the default choice for high-stakes workloads where reliability and support matter more than cost. But the 29% open-weight share is large enough to create a viable alternative ecosystem, which will pressure closed-model margins. The comparison to prior periods is instructive. In April 2024, open-weight models were essentially negligible in production inference. The jump to 11% by April 2025, then 29% by December 2025, represents a compound monthly growth rate of roughly 15%. If that rate holds, open-weight models will hit majority share by mid-2026. Even if growth decelerates, the trend is clear: the open-weight ecosystem has crossed the chasm from hobbyist to enterprise.

Mentioned in this article

Enjoyed this article?
Share:

AI Toolslive

Five one-click lenses on this article. Cached for 24h.

Pick a tool above to generate an instant lens on this article.

Related Articles

From the lab

The framework underneath this story

Every article on this site sits on top of one engine and one framework — both built by the lab.

More in Products & Launches

View all