Skip to content

GPT-image 2.6 Is Now Live on TokenHub โ€‹

Updated

2.6, the latest generation of the GPT-image family, is now available on TokenHub. It uses the existing og-image endpoints โ€” just switch model to one of the two IDs below:

Model IDPositioning
custom-model-og-v2.6-fFlare โ€” faster
custom-model-og-v2.6-sSunburst โ€” more detail

1. About 50% Faster Generation โ€‹

Overall generation is about 50% faster than 2.5.

  • Flare: suited to fast iteration and batch drafts, reaching roughly 6 seconds per image at the low tier.
  • Sunburst: suited to complex compositions and fine-grained editing, with better consistency for characters and products across multiple rounds of revision.

2. New Capabilities โ€‹

CapabilityDescription
Animated GIFoutput_format=gif, supporting 2~16 frames with adjustable per-frame duration (default 9 frames ร— 200ms โ‰ˆ a 1.8-second loop)
SVG vector outputoutput_format=svg, producing infinitely scalable vectors directly โ€” ideal for icons, illustrations, and diagrams
Reference imagesUp to 16 input images when editing
Markup-based editingUse a transparent PNG annotation layer to circle the region to change โ€” far more precise than describing the location in words
complexity decoupled from qualitycomplexity (planning effort before rendering) is controlled independently of quality (pixel precision), with no separate charge

complexity defaults to high. For local edits such as swapping a background or changing a signboard, set it to low to skip re-laying-out the scene.

3. Pricing โ€‹

The token unit price is identical to Image 2.5. However, the output image token count rises significantly at the low and medium tiers, so per-image cost at those two tiers goes up. high and above are unchanged from 2.5.

Taking 1024x1024 as an example, output image token counts compare as follows:

quality2.52.6Change
low196823about 4.2ร—
medium4391,449about 3.3ร—
high1,7561,756Unchanged
xhigh3,1223,122Unchanged
max7,0247,024Unchanged

The token counts above are reference values; actual consumption is whatever the endpoint returns in usage.total_tokens.

If your workload is dominated by low-quality batch generation (icons, thumbnails, bulk drafts), recalculate your budget. At high and above, 2.6 is "same price, faster" and can be adopted directly.

4. Before You Start โ€‹

This model is a test model and requires an allowlist entry granted by the backend team before it can be used. Contact the product team and submit your UIN and AppId. Allowlisting takes effect per model ID, so -f and -s must be requested separately. GIF output additionally requires the model's GIF preview entitlement.

Built with VitePress ยท Deployed on Tencent EdgeOne Pages