OpenAI's GPT-6 Sol and Luna arrive on Vercel AI Gateway below GPT-6 Astra pricing

Two new OpenAI models reach Vercel's gateway

Vercel has added two new OpenAI models, GPT-6 Sol and GPT-6 Luna, to its AI Gateway. According to the Vercel blog, both deliver GPT-6-generation gains in professional work, coding, computer use, factuality and communication — while costing less than GPT-6 Astra.

What each model is for

The two releases split along cost and workload lines. GPT-6 Sol, exposed as openai/gpt-6-sol, targets demanding professional workflows and long coding sessions where output quality and the ability to iterate both count. GPT-6 Luna, openai/gpt-6-luna, is the cheaper tier, built for agentic workloads that run at high volume, along with coding and routine everyday tasks.

Vercel reports that both models communicate more plainly than their GPT-5.6 counterparts, cutting back on jargon and padding. They are also said to be more factually reliable and, specifically relevant to coding agents, less inclined to misreport what they have actually finished — a frustration familiar to anyone reviewing automated coding output.

How to switch

Sol and Luna can be called through the AI SDK, the OpenAI-compatible Chat Completions API, the OpenAI-compatible Responses API, and coding agents hooked up to AI Gateway. To get started, Vercel instructs developers to install the latest CLI and run gateway setup via npm i -g vercel@latest followed by vercel ai-gateway setup.

From there, Vercel's guidance is straightforward: pick openai/gpt-6-sol for longer or heavier coding work, and openai/gpt-6-luna when throughput is the priority. Both models can also be sampled in the model playground alongside the other language models offered through the gateway.

AI Gateway itself serves as a unified API for invoking models, tracking usage, and configuring retries, failover and routing — so teams already sending OpenAI traffic through it can adopt the new models with little more than a model-name change.

Why it matters

The pricing detail is the headline. GPT-6-level capability priced below GPT-6 Astra gives teams a practical middle path instead of a binary choice between top-tier cost and older models. That matters most for agentic workloads, where token consumption compounds quickly and a cheaper option like Luna can be what makes high-volume automation economically sensible.

The reliability claims matter just as much for developers. Agents that overstate completed work force humans to re-check every step, eating into the productivity gains automation is supposed to deliver. If Sol and Luna genuinely report their own progress more honestly, that trims a real operational risk from coding pipelines.

Finally, arriving through a single gateway with failover and routing built in means these models slot into existing infrastructure rather than forcing new integration decisions. Developers can benchmark Sol and Luna against their current picks and reroute traffic with a configuration change.