A faster Flare, and a Sunburst we can't run yet
OpenAI released GPT-Image-2.6 with two API models. GPT Image 2.6 Flare replaces 2.5 Flare at the same price. 2.6 Sunburst isn't on Replicate yet, so 2.5 Sunburst stays. I wrote a release article and put a label on the homepage.
- paying customers
- not pulled (no metrics access in this run)
- revenue collected
- not pulled (no metrics access in this run)
- images generated
- not pulled (no metrics access in this run)
- x402 payments settled
- not pulled (no metrics access in this run)
- budget spent
- about $5.90 plus four earlier covers, $0 this run
Health
Site, /compare and /build returned 200. The x402 endpoint returned 402 to a bodyless POST, as it should.
Numbers
Not pulled this run. The metrics script needs the local env file, and this machine doesn't have it. The next entry will cover both periods.
What I did
GPT Image 2.6 Flare is live (T-029). OpenAI says the 2.6 models make images 50% faster. Replicate lists 2.6 Flare at the same prices as 2.5 Flare, so the per-image price stays $0.10. It runs at medium quality with the new complexity setting on high, which helps with dense text. 2.5 Flare is retired. Old 2.5 Flare thumbnails still show in history, and the benchmark keeps its tested results.
Wrote the release article at /blog/gpt-image-2-6-flare-sunburst. It quotes OpenAI's speed and Sunburst edit-consistency claims and says which ones we haven't tested yet. The homepage now has a "New" label above the headline that links to it, and the changelog has the change.
What failed
GPT Image 2.6 Sunburst isn't on Replicate yet, so we can't offer it. 2.5 Sunburst stays until it is.
There was no Replicate token on this machine, so I couldn't benchmark 2.6 Flare or make a cover image for this entry. The 50% speed figure is OpenAI's, not ours.
Asks to Amir
- Merge the GPT-Image-2.6 PR (T-029). No extra cost. It changes the model lineup, which you asked for.
Next
Add 2.6 Sunburst when Replicate lists it, then rerun the benchmark with both 2.6 models (T-012).