Skip to content
chornous.dev

Type two or more letters. Esc closes.

Index of sheets
Theme

Sheet 03 · Writing

All notes

Grok 4.6 and 4.7 on the API: two frontier models in two months

xAI shipped Grok 4.6 in August and Grok 4.7 in September 2026, with the same 500k context window and the same price. The release notes for developers, with a cost check and a deprecation to schedule.

Pl. 16 · network drawing generated from the slug “xai-grok-4-7-api”

xAI's API release notes list a new frontier model in August 2026, Grok 4.6, and another in September, Grok 4.7. Both carry the same description, a model "for coding, agentic tasks, and knowledge work", and most of the same specs.

context window, 4.6 and 4.7
500k
input / output per 1M tokens, under 200k
$2 / $6
reasoning effort levels, low to xhigh
4

Shared specs

Both take text and image input, return text, and accept a reasoning effort of low, medium, high or xhigh. Both use a two-tier price that depends on prompt length:

Per 1M tokens, from the xAI release notes
Prompt sizeInputCached inputOutput
under 200k tokens$2$0.50$6
over 200k tokens$4$1$12

Design around the jump at 200k. With a 500k window you can paste a whole repository into one request, and past 200k every token in that request costs twice as much. For code work I retrieve the files a task needs and stay under the line. Cached input costs a quarter of fresh input, so keep stable context first and unchanged between calls.

New in 4.7

The release notes give 4.7 three things 4.6 doesn't have:

  • A US regional endpoint, for teams with data-residency rules.
  • A "Grok 4.7 Fast" variant, available only inside Cursor and xAI's Grok Build tool, not on the API.
  • On the Responses API, grok-4.7 always returns reasoning.encrypted_content, even when include doesn't ask for it.

The reasoning effort setting deserves a test on either model. For a code-review bot, low on small diffs and high on large ones keeps latency and cost in proportion to the change. The parameter defaults to high, one step below xhigh, so a bot that leaves it out pays for more reasoning than a small diff needs.

const response = await fetch("https://api.x.ai/v1/chat/completions", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.XAI_API_KEY}`,
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    model: "grok-4.7",
    reasoning_effort: diffLines > 400 ? "high" : "low",
    messages: [{ role: "user", content: reviewPrompt }],
  }),
});

Around the models

  • Grok Bot (August): durable AI teammates that run on a persistent cloud computer, with messaging and connectors.
  • Imagine: an auto quality default, up to five source images for edits, and new 21:9 and 5:2 aspect ratios.
  • Voice: Grok Voice Transcribe 2.0 for speech-to-text. Version 1.0 stays the default.
  • Priority processing (June): a service_tier: "priority" option on text endpoints.

On the calendar: November 2

The grok-imagine-image-quality model retires on November 2, 2026. Its requests will route to grok-imagine-image-2.0 with quality set to low, at a lower per-image price. If your product depends on the old model's output quality, pin the new model and a quality setting yourself before that date.

The release notes list no Grok 5 yet. With two releases a month apart at one price, I put the model id in configuration so the next switch is a one-line change.

Sources

Volodymyr Chornous