TypeSafe AI's Jev becomes Vercel's fastest-adopted AI model
Vercel data shows TypeSafe AI's decision model Jev reached 35% of paid teams within two weeks of launch, nearly double the reach of runner-up Sonnet 5.5.
Jev, a lightweight decision model built by TypeSafe AI, has posted the fastest early uptake of any model tracked by Vercel, according to usage figures released by the cloud platform. Within two weeks of its limited release on September 15, 2026, Jev was in use at 35% of paid teams on Vercel.
That reach is close to twice that of the second most widely used model, Sonnet 5.5. Vercel's figures indicate no model has spread faster during its first fortnight on the platform.
Fewer outputs, lower cost
Jev is built to pick between a small set of predetermined outputs rather than generate open-ended text. Constraining what it can return allows it to run faster and more cheaply than general-purpose models such as Sonnet, which makes it well suited to sorting large volumes of data at speed.
The approach has drawn wide attention since launch and has prompted rival products from OpenAI, Amazon and others. Until now, though, there was little hard evidence that the enthusiasm was turning into real workloads. Vercel itself was an early customer, swapping out GPT-6 Luna in favour of Jev for classifying harmful commands.
The numbers come from Vercel's AI gateway, a unified API service comparable to OpenRouter. That gives only a partial picture: labs such as OpenAI and Anthropic sell access through their own APIs, so their usage may be undercounted in the gateway data.
Open source keeps gaining ground
The report also points to sustained growth for open source models, which now make up more than 70% of tokens processed through Vercel's gateway and nearly 25% of total spending.
The overall market is expanding quickly as well. Spending on frontier labs has risen tenfold even as their share of the total has shrunk, lending weight to the view that frontier models and open source alternatives may complement rather than replace one another.
Comments
No comments yet. Start the discussion or ask a question.
Related
The Granite team released Granite Switch 4.1 on October 15, 2026, combining chat, decision-making adapters and speech-to-text in a single hosted endpoint.
Palo Alto startup Elorian released its first model, Glimpse, on October 16, 2026, pitching it as a system that revisits images and video while it reasons.
Meta has released Muse Spark Open, an 816-billion-parameter open-weight AI model that, after optimisation work with Apple, can run locally on two M5 Ultra Mac Studios.

