Imagine, Voice, and Deprecations: Updating Your Usage
On this page
Grok’s image, video, and voice APIs have evolved as fast as text models, with several removals along the way. This final lesson covers what’s changed since summer and what you need to modify in your code.
Imagine Image 2.0
Released August 7, 2026, Imagine Image 2.0 (grok-imagine-image-2.0) is now the recommended image model. The publisher highlights its fidelity to instructions, typography and layout handling, and targeted editing tools: a magic wand that modifies the designated area without touching the rest, segmentation, cutout with transparent background, and editing from multiple reference images, up to five per request instead of three. Two settings changed in August: the quality parameter accepts auto, which is now the default—it currently uses low for generation and medium for editing, and the image is billed at the served quality—and two new formats appeared, 21:9 and 5:2.
Watch out for a removal effective today: since November 2, 2026, grok-imagine-image-quality is retired. Requests calling it are served by Imagine Image 2.0 at low quality, with no change to request or response shape and at a lower price per image. Your code won’t break, but the output changes: replace the identifier with grok-imagine-image-2.0 and explicitly choose the desired quality. The grok-imagine-image model (version 1.0) is not affected.
| Model | Price |
|---|---|
grok-imagine-image-2.0 | $0.04 per image |
grok-imagine-image | $0.02 per image |
grok-imagine-video-1.5 | $0.080 per second |
grok-imagine-video-1.5-lite | $0.020 per second |
grok-imagine-video | $0.050 per second |
On the video side, Imagine Video 1.5 handles text-to-video, image-to-video, and video from references, with predefined voices optional and native 1080p output for the first two modes. A more economical variant, grok-imagine-video-1.5-lite, is in the catalog.
Voice: Grok Voice 3.0
Speech-to-speech has seen two versions since summer. Grok Voice 2.0 launched in July; Grok Voice 3.0, announced October 14, 2026, replaced it at the same price—$0.08 per minute—and the alias grok-voice-latest now points to it, so sessions using the alias were upgraded without API changes. The publisher announces latency under one second and a better customer service call resolution rate. Grok Voice 2.0 is deprecated: replace grok-voice-2.0 with grok-voice-3.0 or the alias. For transcription, grok-voice-transcribe-2.0 is the default model; version 1.0 reached end of life on October 2, 2026, and its requests are redirected to 2.0 at the same price.
Two New Fields Useful in Production
Since October, a guidelines array lets you adjust, request by request, how Grok handles content in the nsfw, violence, and profanity categories, in Chat Completions and Responses API. Since September, the safety_identifier field lets you pass an opaque identifier for the end user: in case of rule violations, the publisher can attribute them to that user rather than your API key. For an application open to the public, this is a simple protection to put in place.
Official sources: Release notes · Models · Imagine Image 2.0 · Grok Voice 3.0
Key Takeaways
grok-imagine-image-qualityretired since November 2, 2026: switch togrok-imagine-image-2.0and set quality- Imagine Image 2.0:
autoquality default, up to five reference images, 21:9 and 5:2 formats - Grok Voice 3.0 replaces 2.0 at same price;
grok-voice-latestpoints to it - Think about
guidelinesfor moderation andsafety_identifierto isolate user abuse
Test Your Knowledge
1. Which Grok model should you use for text and code as of November 2, 2026?
Answer: Grok 4.7 (grok-4.7), presented by the publisher as its most capable model: 500,000 token context, reasoning effort from low to xhigh, high by default.
2. Your code calls `grok-imagine-image-quality`: what happens since November 2?
Answer: The request is served by Imagine Image 2.0 at low quality, at a lower price and with no format change. The code won’t break, but the output changes: replace the identifier and choose the quality.
3. What is a Grok Bot?
Answer: A permanent agent with its own cloud computer, working in your tools, carrying out a task end-to-end and soliciting you for validations. Team Bots are shared within a team.
4. Does a SuperGrok subscriber need to change plans?
Answer: No. They can keep their current plan or switch to The X Subscription from their billing settings, with a pro-rated refund.
5. What is the `safety_identifier` field for?
Answer: To pass an opaque identifier for the end user, so that a rule violation is attributed to them rather than your API key.