Skip to content

GPT-6.2 Astra and Sol land on Vercel AI Gateway

Vercel added OpenAI's two new GPT-6.2 variants to its gateway, with an ultrafast service tier and a CLI setup step for coding agents.

Draftship Editorial3 min read

Two new models, one endpoint

GPT-6.2 on AI Gateway now covers both of OpenAI's new variants. According to Vercel's changelog entry dated 17 Nov 2026, GPT-6.2 Astra and GPT-6.2 Sol are served through the gateway, so anyone already calling models there gets two more model names rather than a new integration to build.

What separates Astra from Sol

The source splits the two by workload rather than size. GPT-6.2 Astra is meant for complex, multi-step projects: Vercel says it produces slides, spreadsheets and documents that follow your templates and the conventions of your field, and that it is better at implementing and testing complex changes in existing codebases. Its computer use is described as more precise, down to individual parts in CAD and simulation tools.

GPT-6.2 Sol is the general model for everyday work. Vercel says it improves on earlier Sol models for agentic workflows, coding and writing, and that it can turn a written brief into an interactive browser experience, such as a playable game. No example of that is attached.

How you call them

The model names are openai/gpt-6.2-astra and openai/gpt-6.2-sol. They are reachable through the AI SDK, through OpenAI-compatible Chat Completions and Responses APIs, and through coding agents connected to AI Gateway. The changelog shows a streamText call using openai/gpt-6.2-sol with a prompt asking for a one-page project brief from meeting notes.

Both models support OpenAI's fast mode and ultrafast service tier. In the AI SDK you set serviceTier: 'ultrafast' in the OpenAI provider options; in Chat Completions and Responses requests you set service_tier. Ultrafast bills at a higher per-token rate than the default tier, with one stated exception: Ultrafast output on GPT-6.2 Astra costs 60% less than Ultrafast output on GPT-6 Astra. That compares two ultrafast tiers, not ultrafast against default.

For agents, setup is a CLI step. Install the latest Vercel CLI and run npx vercel ai-gateway setup, then select one of the two model names in Codex, Cursor or another connected agent. Vercel also points to a playground for each model.

What it changes for people building apps

Switching models stays a string edit. AI Gateway presents one API for calling models, tracking usage and cost, and configuring retries, failover and performance optimizations aimed at uptime above what a single provider gives you. It also carries custom reporting, Zero Data Retention support, budgets per API key and routing rules. Vercel states that gateway pricing mirrors provider pricing with no markup and no platform fee on inference, including on Bring Your Own Key requests.

This suggests the deciding question is not access but routing: whether Astra's cost per task on multi-step work beats Sol on the same job. The source offers no cost-per-task data, so that comparison is something you would have to measure yourself.

Limits and open questions

The announcement is thin on the numbers a budget depends on. It gives no per-token prices for either model on the default tier, no context window, no latency figures, no rate limits and no benchmark results. The 60% figure is a relative comparison to GPT-6 Astra's ultrafast output only.

It also does not say which earlier Sol models are being improved on, or by how much, and it does not name any models being retired. Region availability for the ultrafast tier is unstated, as is what "fast mode" changes relative to default.

What this means for building apps with AI

More model choice behind one key mainly helps people who build with AI rather than train it. If your app calls a model at runtime, the model name is a configuration detail you can change without rewriting the request layer. Inside an AI app builder, the same idea applies at the tool level: knowing which model handles which job is worth more than chasing the newest name. Draftship's notes on using the AI chat and choosing a model cover that decision from the builder's side. Draftship is not affiliated with Vercel or OpenAI, and nothing here suggests Draftship supports these specific models.

Primary sources

This article was drafted with AI assistance from the primary sources listed below and checked automatically against them. An editor hasn't reviewed it yet.

AI model launches

Gemini 4 Hydrogen lands on Vercel's AI Gateway

Vercel added Google's Gemini 4 Hydrogen to AI Gateway as google/gemini-4-hydrogen, claiming better agentic coding, tool use and grounding than Gemini 3.8 Flash.