OpenAI shipped ChatGPT Images 2.5 on September 8, 2026. PressBot 1.8.7 puts both API models in the admin image catalog the same week: GPT Image 2.5 Flare (gpt-image-2.5-flare, the new default) and GPT Image 2.5 Sunburst (gpt-image-2.5-sunburst). GPT Image 2 stays as the previous-generation option if you already have prompts tuned to it.
These IDs are image-only. They do not appear in public chat or admin text dropdowns. PressBot uses them for generation work, especially featured images in the content pipeline.
What actually improved
OpenAI’s own writeup is the useful source here, not a vibe. Images 2.5 is built around three practical jumps:
Fidelity. Lighting and textures look more natural. Reference subjects hold identity better when you restyle or recompose. That matters if you iterate on a blog visual without wanting a new character every pass.
Precision edits. The model is better at changing only what you asked for. Swap a background, a product, or a caption while keeping composition and the rest of the scene. Earlier image models loved to “help” by redrawing the whole frame.
Multi-turn consistency. Later edits are more likely to keep earlier decisions instead of degrading the asset after three regenerations. For a WordPress featured image, that is the difference between a two-minute tweak and starting over.
OpenAI also reports up to about 50% lower latency versus Images 2.0 on the Flare path, with Flare delivering higher quality than GPT Image 2 at that speed. Treat vendor latency claims as directional until you time it on your own key and host. The qualitative bits (edit locality, subject hold, lighting) are the ones you will notice first in PressBot.
ChatGPT itself got Sketch, templates, and on-image comments. Those are ChatGPT product features, not PressBot UI. In WordPress you still prompt the admin agent. You just get the better API models behind that prompt.
Flare vs Sunburst vs GPT Image 2
Flare is PressBot’s default. Fast, everyday featured images: blog posts, changelog notes, “we shipped this” pieces. OpenAI positions it for high-volume and rapid prototyping. That maps cleanly to a content pipeline that may generate an image on every draft.
Sunburst trades time for tighter control. Use it when the image has to carry a landing page, a cornerstone guide, or a campaign visual you will stare at for months. Longer generation, more precision across edits.
GPT Image 2 remains selectable. If a client site already has a prompt library that looks right on Image 2, do not force a cutover on day one. Generate one Flare and one Image 2 from the same brief, pick a winner, then switch the default.
In PressBot Settings the labels are literal: “GPT Image 2.5 Flare — Fast (Default)” and “GPT Image 2.5 Sunburst — Higher Precision.” Pick it once; the pipeline and agent image tools use that catalog entry.
Cost: same token rates, you still pay OpenAI
PressBot’s catalog notes (checked against OpenAI’s image-prompting guide when 1.8.7 shipped) keep Flare, Sunburst, and GPT Image 2 on the same published per-token rates as GPT Image 2: about $5 / 1M text input, $8 / 1M image input, $30 / 1M image output. Same list price does not mean identical bill per accepted image. Token use and retries still move the number. Time a few real posts before you assume Flare is “free speed.”
You bring your own OpenAI key. PressBot does not add a per-image markup. A better model in the catalog is more capability on the same Pro subscription, plus whatever OpenAI charges you.
Where it sits in the pipeline
PressBot’s content pipeline is still: research → write → critic → optional featured image → categorize → draft. It does not publish. The image step runs after the critic so the visual prompt matches the actual copy, not an outline that changed mid-run.
A typical ask:
Draft a post about GA4 event tracking for WooCommerce. Generate a featured image with Flare. Editorial illustration, no fake UI screenshots, muted palette.
You get a saved draft, preview links, and an image on the post. Then you refine:
Keep the composition. Make the lighting cooler and drop the extra icons in the corner.
That second instruction is the 2.5 pitch. Older models often rebuilt the scene. Flare is supposed to keep the scene and change the lighting. If the result still fights you, switch that post to Sunburst and run one more edit, not a brand-new concept.
The loop is the product: empty editor → draft with image → two tight edits → you hit publish. Faster than briefing a designer for a standard blog header. Not a replacement for a real art director on a flagship campaign.
Agencies feel the clock more than the catalog
One site: convenience. Ten client WordPress installs: calendar time. Same Freemius plugin, your OpenAI key, drafts that already have featured images. Editors stop facing blank screens and spend the hour on local facts, tone, and approval.
Keep it human-in-the-loop on purpose. Auto-publishing across client properties is how you get embarrassed. Pipeline work ends as drafts. Destructive admin actions still need confirmation. Trusted Plans and review-required automations are the same idea for scheduled site work: prepare, you approve, then it applies.
A sane agency default: Flare on every pipeline draft, Sunburst only on pages that stay in the header for a quarter. Do not Sunburst every recap post unless you like waiting and burning tokens.
Models get better; Pro does not ratchet with each drop
PressBot Pro is $29/year. ZIP from pressbot.io via Freemius, not wordpress.org. Bring your own keys (Claude, OpenAI, Gemini, DeepSeek). You pay the providers; we do not skim per message.
When OpenAI ships a better image model, or DeepSeek ships a cheaper Flash, we register it. Your existing Pro install can select it. You still pay OpenAI for pixels. You do not pay PressBot a “GPT Image 2.5 tier.”
We will not pretend PressBot wins every comparison chart. For BYOK WordPress (visitor chat, gated admin agent, draft-first pipeline with real Images API models) it is a strong budget option that gets more capable when the frontier does, without a pricing ratchet tied to each model name.
Try it on 1.8.7
Update to 1.8.7. Set the OpenAI image default to Flare. Run one pipeline draft with an explicit image brief. If the first image is 80% right, ask for a local edit instead of a full regen. Use Sunburst on the next cornerstone piece if you need more hold on details.
Free still covers the visitor chatbot. Pro unlocks the pipeline, image models, and admin tools on /pro/. OpenAI’s announcement is the upstream story; PressBot’s job is to put those IDs in the dropdown the week they ship, then stay out of your way while you approve the draft.
