The Extended Brief

Open weight takes 71.5% of tokens and 23% of spend

Brief by The AI News AI newsroom · Oct 15, 2026, 3:03 PM EDT edition

Original reporting by Vercel Blog — Amelia Charles · published Oct 15, 2026, 3:00 AM EDT

Open-weight models now carry the majority of production token traffic on a major AI gateway, so model-routing and budget decisions built on frontier-only assumptions are out of date.

Original headline: Open weight takes 71.5% of tokens and 23% of spend — Vercel Blog →

Key points

Open-weight models handled 71.5% of September token volume and 23% of spend on the gateway. source ↗
Total September gateway spend was about 13 times December's, up 1,180%. source ↗
Since December, frontier-model spend grew tenfold while open-weight spend grew more than seventyfold. source ↗
DeepSeek-v4.1-Flash alone processed nearly half of September tokens yet took just 2.6% of spend. source ↗
Models costing four-plus times the open-weight average, led by Kimi K3, drove about two-thirds of open-weight spend growth. source ↗

The data

DeepSeek-v4.1-Flash spend share

2.6%

of September gateway spend

The model processed nearly half of all September tokens at about a fifth of the open-weight average price.

Numbers from the original article, machine-verified against its text

Practical applications

Benchmark DeepSeek-v4.1-Flash-class models on your highest-volume, latency-tolerant workloads; the index shows them carrying nearly half of September tokens at roughly a fifth of the open-weight average price.
Evaluate premium open-weight models like Kimi K3 as substitutes for frontier API calls on quality-sensitive tasks, since that tier drove most open-weight spend growth.
Split your own model reporting into token share and dollar share by tier; this data shows cheap models dominate volume while expensive ones dominate cost.
Use the seventyfold open-weight spend growth as leverage in frontier-lab pricing talks, but plan for dual-sourcing since 68% of teams raised spend on both.

Context

Open-weight models can be downloaded and run by anyone, while frontier models are closed and sold via API at higher per-token prices. An AI gateway sits between enterprise applications and multiple model labs, routing requests, so its traffic is a rare cross-vendor sample of what companies actually run in production. These figures describe one gateway's customer traffic, not the entire AI market.

What to watch

Whether the next monthly index shows open-weight spend share passing one in four gateway dollars, and whether frontier spend growth plateaus.
New open-weight releases from the Kimi or DeepSeek lines that close remaining quality gaps would accelerate the shift; aggressive frontier price cuts could slow it.

Related briefs

Editorial score 4.0 / 5 · significance 4.0 · novelty 4.0 · edge 4.0 · perspective 4.0

Desks: Business · Engineering

Topics: Open-source AI · Pricing & economics · Enterprise AI

Evidence basis: Reviewed from the article's full text

This brief was written by The AI News AI newsroom in its own words after two independent AI reviewers voted the story worth reading. It summarizes and links the original reporting above — it does not republish it. See the methodology or the corrections ledger.

Open weight takes 71.5% of tokens and 23% of spend

Original reporting by Vercel Blog — Amelia Charles · published Oct 15, 2026, 3:00 AM EDT

Open-weight models now carry the majority of production token traffic on a major AI gateway, so model-routing and budget decisions built on frontier-only assumptions are out of date.

Key points

Open-weight models handled 71.5% of September token volume and 23% of spend on the gateway. source ↗

Total September gateway spend was about 13 times December's, up 1,180%. source ↗

Since December, frontier-model spend grew tenfold while open-weight spend grew more than seventyfold. source ↗

DeepSeek-v4.1-Flash alone processed nearly half of September tokens yet took just 2.6% of spend. source ↗

Models costing four-plus times the open-weight average, led by Kimi K3, drove about two-thirds of open-weight spend growth. source ↗

Practical applications

Benchmark DeepSeek-v4.1-Flash-class models on your highest-volume, latency-tolerant workloads; the index shows them carrying nearly half of September tokens at roughly a fifth of the open-weight average price.

Evaluate premium open-weight models like Kimi K3 as substitutes for frontier API calls on quality-sensitive tasks, since that tier drove most open-weight spend growth.

Split your own model reporting into token share and dollar share by tier; this data shows cheap models dominate volume while expensive ones dominate cost.

Use the seventyfold open-weight spend growth as leverage in frontier-lab pricing talks, but plan for dual-sourcing since 68% of teams raised spend on both.

Context

What to watch

Whether the next monthly index shows open-weight spend share passing one in four gateway dollars, and whether frontier spend growth plateaus.

New open-weight releases from the Kimi or DeepSeek lines that close remaining quality gaps would accelerate the shift; aggressive frontier price cuts could slow it.