Grok 4.7 rolls into Copilot and reaches 94% in Vercel’s Next.js agent eval

xAI released Grok 4.7 on 21 September as `grok-4.7` on its public API, positioning the model for coding, agentic tasks and knowledge work. The release notes specify a 500,000-token context window, text and image input with text output, and four reasoning-effort settings: low, medium, high and xhigh. High is the default reasoning level.

The same model is also available through Grok Build, Cursor and model gateways. xAI says a faster serving variant runs at roughly twice the output speed and twice the token price, but that Fast variant is limited to Cursor and Grok Build rather than the public xAI API. The standard API release therefore has a broader integration path than the accelerated serving option.

For prompts at or below 200,000 tokens, xAI lists Grok 4.7 at $2 per million input tokens, $0.50 per million cached input tokens and $6 per million output tokens. Above that prompt threshold, the listed rates double to $4, $1 and $12 respectively. That long-context tier matters because the model can accept up to 500,000 tokens even though the cheapest rate applies only to the first pricing band.

xAI also says Grok 4.7 keeps the same base price as Grok 4.6. Its launch post publishes benchmark gains over 4.6 across several coding and professional-work tests, but those measurements are vendor-run. They establish xAI’s launch claims, not an independent performance ranking against other frontier models.

GitHub separately announced that Grok 4.7 is rolling into Copilot for Pro, Pro+, Max, Business and Enterprise subscriptions. GitHub lists Visual Studio Code, Visual Studio, Copilot CLI, Copilot cloud agent, the GitHub Copilot app, JetBrains IDEs, Xcode and Eclipse among the model-picker surfaces receiving it.

The rollout is gradual, so GitHub explicitly warns that eligible users may not see the model immediately. Business and Enterprise administrators can control access through Copilot model policy; under GitHub’s default model-enablement policy, newly released models are enabled unless an administrator has disabled the global default or the model specifically. GitHub says Copilot use of Grok 4.7 is billed at provider list pricing under its usage-based billing system.

Vercel’s Next.js agent eval, last run on 22 September, reports Grok 4.7 with OpenCode at a 94% success rate and $0.109 average list cost per eval. Claude Opus 5.5 at high reasoning with Claude Code, GPT-6 Sol at high reasoning with Codex, and Claude Fable 5.1 at high reasoning with Claude Code each scored 97%, with listed average costs of $0.234, $0.244 and $0.722 respectively. Rauch described Grok as roughly two to seven times cheaper than those three entries; that comparison is scoped to Vercel’s published average list-cost figures for this evaluation.

The table measures model-agent systems rather than model weights in isolation, and the leading entries use different agent harnesses. Vercel says average list cost is estimated from recorded token use at provider public list prices and can differ from an actual bill because of caching, discounts and subscription plans. The page is rerun as the eval set changes, so this is a dated 22 September snapshot rather than a general model ranking or a durable production-cost guarantee.