grok-imagine-image-2.0 price: $0.04 to $0.08 per image, $0.01 input

On xAI's pricing page, grok-imagine-image-2.0 is priced per output image by resolution and quality: $0.04 at 1K with low quality, up to $0.08 at 2K with medium quality. Input images add $0.01 each. xAI's models page headlines the model at $0.04 per image.

Figures are from xAI's pricing and models pages, read 2026-10-02.

What is the price at each resolution and quality?

The model data on the same page lists the supported qualities as low, medium and auto, with auto as the default. The pricing table has no auto row, so if you leave quality unset, check what a test call is billed before assuming a tier.

Is an image edit billed differently from a text-to-image call?

Yes. The models page says image generation uses flat per-image pricing regardless of prompt length, and that image edits are billed for both the input image and the generated output image. The pricing page lists the input charge as $0.01 per image.

Plan edits as a separate line item from generation, because a pipeline that edits many source photos pays the input fee on each one. For one input image and one output, a 1K low edit is $0.01 + $0.04 = $0.05, and a 2K medium edit is $0.01 + $0.08 = $0.09. The page does not spell out how the input fee applies when one request returns several images, so confirm that with a single test call.

How does the older grok-imagine-image row compare?

The same table lists grok-imagine-image at $0.02 at 1K and 2K, and grok-imagine-image-quality at $0.05 (1K), $0.06 (1.5K) and $0.07 (2K). The xAI docs navigation lists "Imagine Image Quality Retirement (Nov 2, 2026)", a page this post did not read; the slug retirement post covers what that means for requests.

What do I pay if I call a Grok image model through Sume?

Sume's Images API docs say the endpoint pricing lines are the amount charged to your wallet, Sume's margin is already applied, and cost_usd × n is what you pay. So xAI's list figures are a reference point, not your Sume price. Read the endpoint record for the model id you intend to call, or run one image and read usage.cost.

Image generation on Sume is all-or-nothing per the same docs: a completed generation is billed in full, and failed or cancelled generations are not billed. That differs from a plain list-price comparison, so budget from the billed usage.cost of real runs.