xAI Docs

Imagine / Images

Imagine overview (/developers/model-capabilities/imagine)

Docs now describe image editing with up to 5 reference images (previously 3).

Multi-image editing (/developers/model-capabilities/images/multi-image-editing)

Limit updated from three to five source images per edit.

Image generation (/developers/model-capabilities/images/generation)

• quality: Allowed values are now low, medium, and auto. When omitted, the default is auto (previously medium). Auto currently uses low for generation and medium for editing; images are billed at the quality served. Parameter remains grok-imagine-image-2.0 only. See quality.
• New aspect-ratio anchors documented for generation (including links to aspect ratio and quality sections).

Developer release notes (/developers/release-notes)

New Imagine image API updates section covering:

• auto quality and default change on grok-imagine-image-2.0
• Up to 5 reference images for editing
• New aspect ratios 21:9 and 5:2 for generation and editing

Audio

Speech-to-text (/developers/model-capabilities/audio/speech-to-text)

For encoding=opus, sample_rate is ignored (Opus packets are sample-rate-agnostic). Docs no longer require specific Opus sample rates; example WebSocket URL drops sample_rate. Encode at the rate your source produces.

REST API — Speech-to-text (/developers/rest-api-reference/inference/speech-to-text)

REST API — Voice (/developers/rest-api-reference/inference/voice)

Query parameter docs updated the same way: sample_rate is ignored with encoding=opus.

Other doc updates

Overview (/overview)

Code API blurb now names the coding model as grok-4.6 (replacing a placeholder).

Examples

System-message examples on text streaming, legacy chat completions, and a sample response on citations now use “helpful and useful” wording instead of “maximally truthful.”