GPT Image 2 deprecated: API removal on February 10, 2027, official replacement is GPT Image 2.6

The gpt-image-2 deprecation is OpenAI's model retirement plan announced on the official deprecations page on 2026-11-10: gpt-image-2 and its snapshot gpt-image-2-2026-04-21 enter deprecated status and leave the API on 2027-02-10, with the recommended replacements gpt-image-2.6-sunburst or gpt-image-2.6-flare named on the same row. It is the third image-model shuffle in the GPT Image family this year, after the DALL·E hard removal in May and the gpt-image-1 shutdown in October. GPTMap fetched the deprecations page, the gpt-image-2 model card, and the pricing page on 2026-11-12; every number and quote below is checked against those fetches.

1. What the notice says

The deprecations page added a 'GPT Image 2' entry under Upcoming deprecations on 2026-11-10 (fetched and checked 2026-11-12). The notice sentence reads:

The gpt-image-2 model and gpt-image-2-2026-04-21 snapshot are deprecated and will be removed from the API on February 10, 2027.

The table has a single row:

Two things to note. First, the deprecated objects are two strings -- the bare model name and the dated snapshot -- so both spellings in your code need to move. Second, the official replacement row points at the two GPT Image 2.6 cards rather than the previous 2.5 generation, consistent with another same-page change: the recommended replacement for the 2026-12-01 shutdown batch has also been updated from 2.5 to 2.6 (see section 4).

2. Model card and pricing page today (fetched 2026-11-12)

The gpt-image-2 model card is still reachable in the model catalog and its status line reads Status: Deprecated; the default snapshot listed on the card is gpt-image-2-2026-04-21, and the snapshot list marks it Deprecated as well. The rest of the card:

• Input modalities text and image, output image; the supported-features list has one entry, inpainting.
• Endpoint support: v1/images/generations, v1/images/edits, and v1/batch are Supported; everything else, including Responses and Chat Completions, is Not supported.
• Standard rate-limit tiers: Build 250,000 TPM / 20 IPM, Launch 3,000,000 / 150, Grow 8,000,000 / 250.

On the pricing page, gpt-image-2 is still listed today: standard rates of $8 image input / $30 output / $2 cached input and $5 text input / $1.25 cached input per million tokens -- row-for-row identical to 2.5 and 2.6; the Batch section carries a half-price gpt-image-2 row (image input $4 / output $15, text input $2.50). Two scope limits, quoted from the page: the footnote states that cached input rates for GPT Image 2 and GPT Image 2.5 apply only to images generated with the Responses API, and the Batch section has no gpt-image-2.6 row as of this fetch -- treat 2.6 Batch billing as unresolved until the page updates.

3. Migration path: swap the model id, keep rates and endpoints

For most callers, migration means changing the model parameter from gpt-image-2 to gpt-image-2.6-sunburst or gpt-image-2.6-flare:

# Scan your repo for the deprecated strings: the bare gpt-image-2 also matches the snapshot
grep -rn "gpt-image-2-2026-04-21\|gpt-image-2" --include="*.py" --include="*.ts" --include="*.json" .

Item by item:

• Endpoints unchanged: both cards support the same three endpoints (generations, edits, Batch), so the HTTP surface of your code does not move.
• Rates unchanged: standard rows are identical cell for cell (numbers in section 2), so migration itself does not change unit cost; cost movement comes from how you use the new 2.6 parameters (complexity tiers, GIF frame budgets).
• New capabilities: the official changelog entry of 2026-11-04 -- GPT-Image-2.6 Sunburst and Flare launched for image generation and editing with adjustable complexity, SVG generation, and animated GIF generation; markup-guided editing extended to supported GPT-Image-2.5 and GPT-Image-2.6.
• Time-boxed window: gpt-image-2 keeps working at current rates until 2027-02-10, but new projects and in-development features are better served landing on 2.6 directly to avoid a second migration.

4. Notice period and the family timeline

Notice math: from the 2026-11-10 announcement to the 2027-02-10 shutdown, 92 days (calendar-computed). The same page lays out minimum notice tiers -- at least 6 months for generally available models; at least 3 months for specialized variants of GA models and image generation models; preview models may get roughly 2 weeks. gpt-image-2 falls in the image generation tier, so 92 days meets the floor, with OpenAI reserving an exception for safety or compliance timelines.

In the context of the GPT Image family's 2026 timeline, this notice closes a chain:

5. Self-check list

Work through these in order:

1. Code and config: grep for gpt-image-2 (it also matches the snapshot string) and change each hit to one of the two official replacement ids; sweep environment variables, IaC templates, and embedded mobile configs too.
2. Batch jobs: make sure no gpt-image-2 batch is scheduled after 2027-02-10; current Batch pricing row status is in section 2.
3. Prompt templates: prompts tuned for gpt-image-2 (quality tiers, size combinations) should be re-validated on 2.6 with a small batch -- the new complexity parameter participates in shaping results, so templates are not guaranteed to transfer verbatim.
4. Snapshot-pinned calls: gpt-image-2-2026-04-21 shuts down with the bare name; do not fix one and miss the other.
5. Post-migration check: re-verify the replacement ids against the deprecations page spelling (gpt-image-2.6-sunburst / gpt-image-2.6-flare); the model ids carry no latest suffix.

6. Common mistakes and troubleshooting

• Migrating the bare name but missing the snapshot: the notice row lists both strings; changing only one leaves snapshot-pinned paths failing.
• Assuming cached pricing is universal: the pricing footnote limits cached input rates for GPT Image 2 and 2.5 to images generated with the Responses API -- re-read the footnote before it enters a cost model.
• Writing 2.5 as the replacement for the 2026-12-01 batch: the official row now recommends 2.6 (it still said 2.5 when GPTMap checked on 2026-10-31) -- migration docs should use the replacement ids from the deprecations page as fetched that day.
• Reading the notice as an immediate availability change: between the notice and the shutdown date, gpt-image-2 remains callable at current rates; live availability is the platform's call -- GPTMap holds no API key and cannot test account-side status.

7. Next steps

• Capabilities and parameters of the migration target: GPT Image 2.6 guide: adjustable complexity, SVG and animated GIF output with Sunburst and Flare.
• Generation and editing patterns for existing gpt-image-2 code: GPT Image 2 API in Practice: From Text-to-Image to Image Editing.
• Cross-vendor image model selection: GPT Image 2 vs Midjourney vs DALL·E 3: Three Image Generation Models in Practice (2026).
• The running timeline of GPT Image family launches and retirements: OpenAI Models Release Notes (2026, Living Document).