Open-Weight AI Models Captured Majority Traffic in August

Open-weight models saw a significant surge in usage as average token prices dropped by over 23 percent last month.

Updated on Oct. 2, 2026 in Artificial Intelligence

Isometric editorial illustration of dense metallic server cooling fins and copper heat pipes in a muted color palette. — Open-weight AI models captured 56% of token volume on the Vercel AI Gateway in August 2026, as developers shifted toward lower-cost alternatives. AI Illustration. Upload story photo >

Live Poll

Do you trust that newer, open-source AI tools are as reliable as expensive proprietary models?

In August 2026, open-weight AI models accounted for 56% of token volume on the Vercel AI Gateway, marking a rapid rise from 7% in December 2025. This shift toward lower-cost alternatives contributed to a 23.2% decline in average token prices across the gateway.

Why it matters

Production users are increasingly shifting toward lower-cost AI models to reduce total token expenditure without sacrificing output. This transition highlights a broader trend where developers prioritize cost-efficiency by migrating workloads from premium models to more economical options.

Open-weight models captured 14% of total gateway spending in August 2026, while high-volume teams saw a 7.6% reduction in per-token costs compared to July. Additionally, models introduced within the last three months accounted for 50% of all token traffic.

The players

Anthropic

This AI research and development company captured 64% of total spending on the Vercel AI Gateway in August 2026.

OpenAI

This artificial intelligence organization launched the GPT-6 Astra model, which quickly captured significant market spending.

Vercel AI Gateway

This platform serves as a routing service that tracks token traffic and spending data for various AI models.

The details

Users migrated workloads from the higher-cost Fable 5 model to the lower-cost Opus 5 model, which saw its share of gateway spend rise to 22.5% as Fable 5 dropped to 4.9%. Meanwhile, Anthropic secured 64% of total gateway spending, and OpenAI’s GPT-6 Astra captured one-third of all dollars spent on OpenAI models within 48 hours of its launch.

December 2025: Open-weight models represented 7% of token traffic.
April 2026: Open-weight models reached 13% of total token volume.
August 2026: Open-weight models became the majority of token traffic.
September 4-16, 2026: GPT-6 Astra and GPT-5.6 Sol processed 27% of OpenAI tokens.

The Tech Race

This migration reflects an industry-wide transition where developers favor open-weight alternatives to replace reliance on legacy proprietary models. The shift follows the Vercel AI Gateway token traffic patterns, which demonstrate a rapid acceleration in the adoption of cost-efficient AI architectures.

The transition to open-weight models allows developers to maintain output while significantly reducing the costs of AI-integrated applications. Users may experience more cost-effective service as companies optimize their workflows by switching to lower-priced model options.

The takeaway

Developers are increasingly leveraging open-weight models to maintain performance while significantly cutting operational overhead. This shift suggests that cost-optimization has become a primary driver in how organizations select and deploy artificial intelligence tools.