Grok 4.7 arrives on Vercel's AI Gateway with 500K context window and 40% launch discount

According to a Vercel blog post, Grok 4.7 — the latest model from SpaceXAI — is now available through the company's AI Gateway, and requests to it are 40 percent off through September 27. The discount is applied automatically whenever the model is called under its gateway identifier, spacexai/grok-4.7, so no promotional codes are needed.

A 500K context window and four reasoning levels

Vercel says the model accepts up to 500,000 tokens of context, which puts it among the larger-context options reachable through the gateway. It also exposes four reasoning tiers — low, medium, high and xhigh — so developers can decide per request how much deliberate computation the model spends before answering. Lower tiers favor fast responses, while xhigh is aimed at harder problems where slower, deeper reasoning pays off.

One identifier across SDKs, APIs and coding agents

The same model ID works everywhere Vercel exposes inference: the AI SDK, the OpenAI-compatible Chat Completions endpoint, and coding agents connected to AI Gateway. For agent use, Vercel's instructions are to install the latest Vercel CLI, run the gateway's setup command, and then select spacexai/grok-4.7 from within supported tools such as fx, Cursor, Codex, Amp and OpenCode.

Vercel's eve agent scaffold can also be initialized directly against the new model, with the reasoning level chosen at creation time — for example, a new agent pinned to Grok 4.7 running at xhigh reasoning.

Pricing, discounts and gateway economics

The launch discount runs through September 27 and applies automatically. Beyond the promotion, Vercel emphasizes that AI Gateway passes provider pricing through without markup and charges no platform fee on inference, including on Bring Your Own Key requests where customers supply their own provider credentials.

For background, AI Gateway is Vercel's unified layer for calling language models. According to the company, it handles usage and cost tracking, retries, failover and performance optimizations intended to deliver uptime beyond what any single provider offers, along with custom reporting, per-key budgets and routing rules.

Privacy and observability

Grok 4.7 supports Zero Data Retention and an option to disallow prompt training, which matters for teams whose compliance rules prohibit prompts or outputs being stored or used to train provider models. Requests to the model also surface in AI Gateway's existing tooling — logs, custom reporting and budgets — next to the other models an organization already routes through the gateway, so adopting Grok 4.7 does not create a separate billing or observability silo.

Why it matters

The addition gives developers already standardized on Vercel's gateway same-day access to a newly released model with an unusually large context window, and the single model ID means switching is close to a one-line change in most integrations.

The tiered reasoning levels are the more strategically interesting feature. Latency and token cost are the main levers teams pull when balancing answer quality against spend, and per-request control over reasoning depth makes that tradeoff explicit rather than baked into the model choice. A team can run cheap, fast calls for routine work and reserve the top tier for genuinely hard tasks.

The enterprise controls — zero data retention and no prompt training — broaden where the model can responsibly be deployed, and no-markup pricing combined with the temporary 40 percent cut lowers the barrier to benchmarking it against incumbents. The catch is timing: the discount expires September 27, leaving a short evaluation window for anyone who wants to test the model at reduced cost.