GPT Image 1.5 Selection and Migration

Updated: 2026-08-21. Model IDs, availability, sunset plans, and billing for 1.5 / 2 follow OpenAI Platform and ChatGPT that day. This article does not invent a concrete retirement date.

Introduction

GPT Image 1.5 was the previous workhorse on OpenAI’s image line. The common API id is gpt-image-1.5 (verify in docs that day). After GPT Image 2 (gpt-image-2) became the flagship, 1.5’s job shifted from “default first pick” to compatibility, A-B checks, cost-sensitive work, and controlled migration. This guide covers when 1.5 still fits, how to plan a move to 2, and how to think about quality versus bill—without announcing a sunset day for OpenAI.

What this guide solves

• Decide whether today’s task should stay on 1.5—or stop chasing flagship by habit
• Verify availability and sunset notes in docs instead of secondhand dates
• Run an evaluate → side-by-side → switch default → retire old ID checklist
• Frame trade-offs across cost, quality, and rework

When 1.5 still makes sense

Real situations where 1.5 fits:

Task signal	Why stay on 1.5
Pipelines / scripts hard-code gpt-image-1.5	Keep reproducibility first, then plan the ID swap
Need visual A-B against historical batches	Same model for a fair comparison
Compliance or client contracts name the old model	Get written approval before changing
Entry only exposes 1.5 for now	Follow the live model list on that entry
Internal drafts and quota-sensitive work	Probe on 1.5; evaluate 2 for finals

Do not force 1.5 for: new external key-art finals, posters that depend on on-image text, complex multi-round local edits, or deliveries where you already confirmed 2 is clearly steadier. Those should move to GPT Image 2 now or soon.

One-liner: 1.5 is the compatibility and A-B tier; 2 is the new-delivery flagship default. Family map: What is GPT Image?.

Sunset and availability: docs only—no invented dates

Whether a model can still be called, is marked deprecated, or is snapshot-only is entirely per OpenAI docs and console that day.

Practical rules:

1. Open the matching model page and changelog on platform.openai.com; record the date you checked.
2. If docs show deprecation / sunset / legacy language, paste the official link and date into team standards—do not put blog rumor dates into contracts.
3. This article cannot promise “offline on month/day X”; when unsure write “follow docs that day” and calendar a biweekly recheck.
4. ChatGPT UI labels may differ from API IDs; migrate against API model ID and the account’s actual selectable list.

Migrate to GPT Image 2: checklist

Do it in order—avoid a global swap with no rollback.

Phase A: Inventory

• List every image entry: ChatGPT / third-party studio / your API
• Record current model field or UI selection (is it gpt-image-1.5?)
• Export representative prompts, reference strategy, size and quality settings
• Mark assets that must stay pixel-consistent (these may stay on 1.5 snapshots longer)

Phase B: Side-by-side (same prompt, dual model)

Pick 10–20 real tasks (heroes, edits, short titles, multi-ref). Run on 1.5 and 2, changing only model, keeping other params as equal as possible.

Score columns suggestion: task | composition | detail | text | edit follow | shippable | notes. Use your acceptance brief—not “looks newer.”

Phase C: Switch default

• New projects default to gpt-image-2 (or the matching flagship UI name)
• Config / env allows per-task override back to 1.5
• Docs and design standards updated: new delivery on 2; historical A-B on 1.5
• Watch failure rate, retries, and time-to-usable-image for one week

Phase D: Retire

• Confirm no active pipeline hard-depends on 1.5 before tightening defaults
• If OpenAI announces deprecation, follow the docs migration window; window dates cite official pages only
• Archive: migration report + side-by-side samples + docs URL and check date

Generation/edit detail for 2: GPT Image 2 complete guide. From-zero ops: Getting started.

Cost vs quality: decision frame

Price, quotas, and rate limits follow docs and console that day. Here is only a decision frame.

Strategy	Approach	You gain	You lose
All 1.5	Every version pinned to the old ID	Predictable behavior; easy historical A-B	May miss 2’s text/edit upside
All 2	Every new task on flagship	Higher per-frame ceiling usually	Harder to explain “why unlike last week” without A-B
Recommended mix	1.5 for old batches; 2 for new delivery; dual-run critical frames	Controllable risk; often less total rework	Needs one selection and config pass

What “cheaper” really means: compare cost to a shippable frame, not one generate price. If 2 saves 2–3 redo rounds, total bill and calendar often still beat “cheap but endlessly repaired” 1.5. Log model, prompt summary, and pass/fail for two weeks—let your data decide.

On quality: 1.5 can still cover internal drafts; external brand key art, short titles that must be correct, and complex multi-round edits should accept on 2. Size and quality enums may not match across generations—reread the day’s parameter tables; do not assume same name = same meaning.

Done checklist

• Acceptance line clear: historical A-B vs new external final
• If choosing 1.5: reason logged (compatibility / contract / A-B), not “habit”
• Platform / ChatGPT checked that day for 1.5 and 2 availability (including deprecation hints)
• Side-by-side samples archived; default switch has a rollback switch
• New-delivery prompts and checklists follow the 2 workflow (composition first, single-point edits)
• Text, logos, and legal copy still zero-tolerance final-check or later typeset
• Bills and time reviewed as probe rounds + refine rounds + model (unit price that day)
• External promises do not hard-code sunset dates, unit prices, or expired size/quality lists

Fast path

• China convenience (third-party): GPT Image 2
• Text-to-image studio: Text-to-image studio
• Official ChatGPT: chatgpt.com
• OpenAI Platform: platform.openai.com

Third-party and official accounts, quotas, and model lists can differ; bind every selection conclusion to a concrete entry. China notes: China access guide.

FAQ

Is 1.5 already retired?

Follow the docs on the day you open them. If gpt-image-1.5 is still listed and your account can call it, controlled use is fine. Whether new projects should keep choosing it is separate—new delivery should evaluate 2 by default.

Exact day of shutdown?

This article invents no date. Check OpenAI Platform model pages and official announcements; store official dates and links in your wiki.

Will old images “change style” after migration?

Possibly. Same prompt on a new model is not pixel-identical. For consistency: keep key 1.5 history frames, or constrain 2 with references—do not assume silent upgrades are invisible.

Can API stay on 1.5 while ChatGPT uses 2?

Yes, but the team must document which pipeline binds which ID so design comps and automation do not diverge. Entry differences: China access.

How big is the price gap?

Not hard-coded here. Open Platform pricing and account usage; compare total cost to shippable, not list price alone.

Official resources

• ChatGPT
• OpenAI Platform
• OpenAI.com

Further reading

• GPT Image guides hub
• What is GPT Image? Family explainer
• GPT Image 2 complete guide
• Getting started: first image
• GPT Image API & Platform

Summary

GPT Image 1.5 can still be a sound choice for compatibility, A-B, and controlled transition, but new external delivery should default-evaluate GPT Image 2. Sunset and price cite official docs that day only; switch defaults with side-by-sides and a migration checklist; decide with total cost to shippable—not single generate price.