$ cat wiki/models/grok-imagine-image-2.md
Grok Imagine Image 2.0
Spec
| Attribute | Value |
|---|---|
| Developer | xAI |
| Released | 2026-08-07 |
| Announced | 2026-08-07 |
| Context window | unknown |
| Pricing | unknown |
| License | unknown |
| Availability | grok.com/imagine (Quality Mode), iOS and Android apps — no API at launch |
Pricing and License are unknown because **no first-party page was readable | |
this run** — the egress proxy blocked x.ai along with every other host — and no | |
| third-party source read carried a figure. The API absence is itself reported rather | |
| than inferred (source). |
Release Date
2026-08-07, generally available on release rather than as a preview — unlike Grok Imagine Video 1.5 (Preview), which shipped as an API preview (source).
Benchmarks
xAI's own claim, as relayed by reporting — no leaderboard was read this run.
| Arena | Claimed rank | Model above it |
|---|---|---|
| Arena text-to-image | #2 | gpt-image-2 (OpenAI) |
| Arena image-edit | #2 | gpt-image-2 (OpenAI) |
| Claimed as of 2026-08-07, with xAI's entries listed as `grok-imagine-image-2 | ||
(low)andgrok-imagine-image-quality` | ||
| (source). |
Two cautions the sources do not resolve. The rank is xAI's, relayed by
third-party coverage, not an independent reading. And the phrase used is "both major
image arenas" without saying which — this repo's own leaderboard snapshots under
sources/evals/ cover LMArena's text tracks and Artificial Analysis, and neither
carries an image-edit column, so there is nothing local to check the claim against
(source).
Use Cases
Editing is the release, and the feature list is an editing tool rather than a prompt-and-hope generator (source):
- Magic wand — changes only the region the user points at, leaving the rest.
- Segmentation — selects precise areas to modify.
- Background removal — exports a subject with transparency for downstream use.
- Multi-reference editing — up to five input images in one generation, removing the manual compositing step that previously meant stitching sources by hand.
- Smart resize across nine aspect ratios.
Compared To
| Model | Provider | Modality | Claimed arena position |
|---|---|---|---|
| Grok Imagine Image 2.0 | xAI | text-to-image + editing | #2 both tracks (xAI's claim) |
| gpt-image-2 | OpenAI | text-to-image + editing | #1 both tracks (per xAI's claim) |
| Grok Imagine Video 1.5 (Preview) | xAI | image-to-video | #1 AA Video Arena I2V (Jun 2026) |
| xAI now claims a top-two placement in image and a top placement in video, from a | |||
| product line that was an API preview two months ago. The comparison worth holding is | |||
| against its own sibling rather than against OpenAI: the video model shipped | |||
| API-first at $0.08/second, and the image model shipped **app-first with no API | |||
| at all** (source). |
Open Questions
- Which arenas? "Both major image arenas" is unresolved in every source read.
- No model card, parameter count, architecture or training detail was published.
- Does Image 2.0 share the Aurora engine that Grok Imagine Video 1.5 (Preview) is built on? No source read says either way.
- When does the API land, and at what price? The absence at launch is reported; a timeline is not.