deniz.in

Markets

Weather

Loading weather

· via Vercel blog

OpenAI's GPT-6.2 Astra and Sol coding models arrive on Vercel AI Gateway

Vercel's AI Gateway now routes OpenAI's GPT-6.2 Astra, aimed at complex multi-step codebase work, and GPT-6.2 Sol for everyday coding, with Astra's Ultrafast output tokens priced 60% below GPT-6 Astra's.

OpenAI's GPT-6.2 Astra and Sol coding models arrive on Vercel AI Gateway

Two models split by workload

OpenAI's newest coding models, GPT-6.2 Astra and GPT-6.2 Sol, are now available through Vercel's AI Gateway, according to an announcement on the company's blog. The pair covers two distinct points on the workload spectrum: Astra is built for complex, multi-step projects, while Sol is meant for everyday coding, writing and agent workflows.

According to Vercel, GPT-6.2 Astra gets better at implementing and testing changes inside existing codebases, at following document templates, and at computer use. That profile points it at teams working in large, established repositories, where the hard part is not writing new code but safely modifying and validating what already exists.

GPT-6.2 Sol is positioned differently. Vercel describes it as suited to everyday coding, writing and agent workflows, including building interactive web applications from a written brief — the kind of quick, self-contained generation work that makes up much of routine development.

Access, APIs and pricing

Both models are reachable under the IDs openai/gpt-6.2-astra and openai/gpt-6.2-sol, with requests authenticated by setting the AI_GATEWAY_API_KEY environment variable to an AI Gateway API key. The two models also support the Responses API.

For latency-sensitive workloads, Vercel says both models support fast mode and OpenAI's Ultrafast service tier, which delivers lower latency at higher per-token rates. In the AI SDK, Ultrafast is requested by setting providerOptions.openai.serviceTier to 'ultrafast'; in Chat Completions and Responses requests, the equivalent is service_tier set to 'ultrafast'.

On cost, Vercel notes that GPT-6.2 Astra's Ultrafast output-token price is 60% lower than that of GPT-6 Astra — a substantial drop for high-volume agent workloads, where generated output tends to dominate the bill.

Wiring the models into coding agents

Vercel also highlights integration with agentic coding tools such as Codex and Cursor. The setup path is to install the latest Vercel CLI and run npx vercel ai-gateway setup, then select openai/gpt-6.2-astra or openai/gpt-6.2-sol from within the agent. Developers who want to evaluate the models before committing can try them in Vercel's playground.

Why it matters

Two things stand out in this release. The first is OpenAI's continued segmentation of its lineup: rather than a single general-purpose model, developers now choose between a heavyweight tuned for navigating and modifying existing codebases and a lighter model for routine generation and agent tasks. That means matching the model — and its cost profile — to the job at hand instead of paying for capability that goes unused.

The second is pricing. A 60% cut on Astra's Ultrafast output tokens relative to GPT-6 Astra signals that fast, high-throughput inference for coding agents is getting materially cheaper. Because agentic coding burns output tokens quickly, generating diffs, tests and explanations along the way, that reduction can meaningfully change the economics of running agents at scale.

Finally, the fact that both models are available through Vercel's AI Gateway, with a single setup command for tools like Codex and Cursor, reflects how gateways are becoming a default distribution channel for frontier models. Developers get one API key and one routing layer across providers, which lowers the friction of trying new models — and switching between them — as they ship.

  • #openai
  • #vercel
  • #ai-gateway
  • #coding-agents
  • #developer-tools

Related posts