Model comparison ยท 2026-09-22

Grok 4.6: when to stay on it while Grok 4.7 is available at the same price

Covers that xai/grok-4.6's catalog description states it's now succeeded by Grok 4.7 at the same price, and the cases where staying on this identifier still makes sense.

Diagram showing Grok 4.6 succeeded by Grok 4.7 at the same price, but still an option for integrations that need stable behavior.

The model page states the successor plainly

xai/grok-4.6's catalog description states it's a strong option for coding, agent flows, and heavy research work, but is now succeeded by Grok 4.7 at the same price. Since the price stayed the same ($2 input / $6 output per million tokens, $4 / $12 at long context), there's no direct cost reason not to choose Grok 4.7 for a new integration.

Still being listed on the model page and remaining public means existing integrations can keep working with this identifier; this repo follows a retirement/migration policy, not removal, for models.

When staying on this identifier is justified

If you already have a production integration tested against Grok 4.6, with verified and known behavior, moving to 4.7 just because it's 'newer,' without a functional need, adds an unnecessary regression-testing burden. Grok 4.6 carries four reasoning_effort levels โ€” low, medium, high, and xhigh; if you've already verified this range gives the behavior you expect in your integration, keeping the version stable is a reasonable choice.

If you're building a new integration, though, the case for deliberately starting with 4.6 instead of the more current model (Grok 4.7) at the same price is weak; starting directly with the current identifier removes the need for a migration project later.

  • A production integration already tested against 4.6: don't move without a functional need.
  • A new integration: starting with the current identifier at the same price makes more sense.
  • If the four reasoning_effort levels (low/medium/high/xhigh) behavior is already verified, keep it fixed.

The data-retention note applies here too

As with other xAI models, this model page also states that the xAI API can retain requests/responses for 30 days for audit purposes; weigh this note separately for a flow handling sensitive data.

Frequently asked questions

Is Grok 4.6 cheaper than Grok 4.7?

No, the catalog description states they're at the same price; the identifier choice should be made on behavior stability, not cost.

Will Grok 4.6 be removed in the future?

This repo follows a retirement/migration policy, not removal, for models; if anything changes it reflects in the model page and status label โ€” follow the current state on the model page.

Related posts