Skip to main content
2026/8/13 01:08 (UTC+8) · New Model · xAI 🚀 Grok Imagine 2 image models are live on the Default group, with 2K landing near 64% of list xAI’s second-generation image model shipped 7 August (officially Grok Imagine Image 2.0), and what we integrate is the official-transit Quality Mode. The two variants grok-imagine-image and grok-imagine-image-quality share one set of endpoints and parameters, differing only in output fidelity and price. Note the model IDs contain no 2 — do not write grok-imagine-2-image. Pricing is the headline here: xAI charges the quality tier by resolution ($0.05 at 1K, $0.07 at 2K), while we charge a flat $0.045 for both — so the higher the resolution, the more you save, landing at about 90% of list at 1K and 64% at 2K. Stack the top-up bonus and the common $100 tier reaches roughly 58%, or 54% at maximum. The standard tier at $0.02/image matches xAI’s list price. From roughly 220 real calls: across 5 aspect ratios x 2 resolution tiers, output pixels matched the request 20/20 exactly (16:9 at 2K reaches 2816x1584); n accepts 1–10; 1K takes about 9 seconds and 2K about 15–17, with 100 RPM running comfortably. Reference editing is genuine editing — only the specified part changes while everything else is preserved — and fusion accepts 1–4 references, with each added image contributing its own subject in testing. Two integration conventions matter, because following the upstream vendor’s documentation will return 400:
  • Image editing must use /v1/images/edits with multipart/form-data file upload; JSON always returns 400 (the upside: no image hosting needed, just send the local file)
  • Reference images must not go to /v1/images/generations — that returns 200 with a normal image, but the reference is silently discarded and you are still billed, with no error at all
Note too that the editing endpoint’s output dimensions follow the first reference image; resolution and aspect_ratio have no effect there. Full parameters, migration guide and code samples: Grok Imagine 2 documentation; measured details in the launch notes. Teams already on GPT-Image-2 can jump straight to the migration section — the endpoints are identical, but the parameter system is not.
← Back to Live Updates · 📚 Monthly archive