What Is GPT Image 2.6? Specs and Markup Edits

What Is GPT Image 2.6? Transparent PNG Markup Overlays and API Specs

GPT Image 2.6 is an unreleased image generation and editing model developed by OpenAI that introduces native support for transparent PNG markup overlays. The model first surfaced publicly on October 28, 2026, inside OpenAI's official OpenAPI specification, which documented an expanded POST /images/edits endpoint. Positioned as the direct successor to GPT Image 2.5, the system interprets freehand sketch annotations and visual guidance layers alongside traditional natural language prompts. While OpenAI has not yet published an active model identifier or pricing schedule, official schema references confirm that the engine is built for precision spatial steering.

Source: @Mr_Salio

Key Takeaways

• Official Specification Discovery: GPT Image 2.6 was identified on October 28, 2026, at 20:56 UTC within OpenAI's openai-openapi repository.
• Markup-Guided Editing: The endpoint introduces a dedicated markup parameter that accepts transparent PNG files to direct visual alterations.
• Dual-Model Support: The API documentation explicitly notes that the new editing mechanism "Supports GPT Image 2.5 and 2.6."
• Mutual Exclusivity Constraint: In the official API schema, requests cannot combine the markup parameter with a standard binary mask.
• Deployment Status: As of late October 2026, GPT Image 2.6 does not possess a callable model identifier, public pricing tier, or confirmed launch date.

What Is GPT Image 2.6?

GPT Image 2.6 represents the next planned iteration in OpenAI's closed-weights visual model line. The family lineage began with the deployment of GPT Image 2 in April 2026, followed by the release of GPT Image 2.5 snapshots—specifically gpt-image-2.5-sunburst and gpt-image-2.5-flare-2026-09-08—in September 2026. Following the official retirement of the legacy gpt-image-1 endpoint on October 23, 2026, OpenAI's image stack transitioned entirely to second-generation architectures.

The first public breadcrumb for version 2.6 appeared when developer trackers, including AI Tracker Bot and Salio, spotted updates to the official OpenAPI definition for the image editing route. Rather than arriving via a formal research paper or staged press release, the model was documented directly within production schema definitions.

Unlike conversational models such as GPT-6.1 Terra or GPT-6.1 Sol, which focus on reasoning and agentic text execution, GPT Image 2.6 is specialized strictly for visual synthesis and localized image manipulation. Its primary technical objective is solving spatial ambiguity: natural language descriptions often struggle to specify exact coordinates, bounding boxes, or directional changes without tedious trial and error. By pairing language with visual markup layers, GPT Image 2.6 enables precise compositional edits.

GPT Image 2.6 at a Glance

Specification Attribute	Details	Verification Source
Developer	OpenAI	Official OpenAPI Repository
Model Type	Multimodal Image Generation & Inpainting	Official OpenAPI Repository
Primary Modality	Text-to-Image, Image-to-Image, Markup Editing	API Schema Documentation
Predecessors	GPT Image 2, GPT Image 2.5 (sunburst, flare)	Official OpenAI Model Index
Key Capability	Markup-Guided Editing via Transparent PNG	OpenAI Image Edits Specification
Context / Input Format	Multipart Form / JSON (File ID, HTTPS URL, Base64)	API Schema Documentation
Mutual Exclusivity	Incompatible with binary mask parameter	API Schema Documentation
Pricing	Not yet confirmed	Awaiting Official Pricing Sheet
Release Status	Unreleased (Referenced in API Spec)	Public Developer Trackers
License	Proprietary / Closed Weights	Vendor Model Architecture

How GPT Image 2.6 Works: Markup Overlays and Editing Mechanics

The defining feature associated with GPT Image 2.6 is the introduction of Markup-Guided Editing. Traditional programmatic image editing in machine learning workflows relies heavily on binary alpha masks. Under an inpainting mask workflow, white pixels indicate regions the model is permitted to redraw, while black pixels remain frozen. This mechanism enforces strict boundaries, but it communicates zero stylistic or structural intent regarding what should appear within the cleared space.

GPT Image 2.6 introduces a functional Transparent Overlay paradigm. The model reads a secondary image file containing colored visual sketches—such as hand-drawn circles, bounding arrows, replacement color fills, or rough object outlines—placed directly over the original canvas.

The core architectural constraint documented in the specification is mutual exclusivity: the markup property cannot be combined with a standard mask in a single edit request. Developers must choose between binary masking and markup guidance.

The API specification outlines several strict structural rules for utilizing this mechanism:

1. Dimensional Matching: The markup overlay must match the exact dimensions of the base image after orientation corrections have been applied.
2. Alpha Channel Requirement: The overlay file must be a PNG format that explicitly contains transparent pixels. An opaque overlay will fail validation.
3. Payload Flexibility: Within JSON requests, the markup field accepts an uploaded file ID, a direct HTTPS URL, or an inline base64-encoded data URI. It is also supported via standard multipart form uploads.

{
  "model": "gpt-image-2.6",
  "image": "file-abc123baseImage",
  "markup": "file-xyz789transparentOverlay",
  "prompt": "Replace the circled storefront with a modern cafe and change the sign text to match the red sketch"
}

Through this mechanism, GPT Image 2.6 shifts the programmatic image editing paradigm from binary alpha masks to semantic, multi-color overlay sketches. The model interprets the position, geometry, and color of the markings on the transparent layer, translating them into generative instructions that align with the text prompt.

What Developers Can Do With GPT Image 2.6

Public discussions and API documentation outline several practical developer workflows enabled by GPT Image 2.6:

• Art Director Annotations: Instead of writing complex spatial prompts (such as "in the upper-left quadrant, roughly 20 percent from the margin"), developers can build client applications where users draw red or blue circles directly over the canvas. The model processes the visual mark as an explicit spatial anchor.
• Rough Compositional Inpainting: By sketching simple line drawings—such as drawing a rough rectangular archway or window frame onto a blank wall—users provide geometric boundaries that the model converts into photorealistic textures.
• Palette and Swatch Steering: An overlay featuring a solid block of translucent teal over a piece of clothing signals the intended color tone to the model, eliminating ambiguous color-bleeding common in pure text editing.
• Directional Repositioning: Drawing arrows to indicate light direction or object movement allows the underlying model to understand dynamic vector instructions that text alone struggles to convey.

How GPT Image 2.6 Compares

To understand where GPT Image 2.6 fits into OpenAI's visual generation timeline, developers must examine the evolution from GPT Image 2 to the current production generation.

Feature / Metric	GPT Image 2	GPT Image 2.5	GPT Image 2.6
Launch Window	April 2026	September 2026	Unreleased
Current Status	Active Base Model	Production Snapshots (sunburst, flare)	Pre-Release Spec Reference
Standard Inpainting Mask	Supported	Supported	Supported
Markup Overlay Support	Unsupported	Supported in Spec Update	Supported in Spec Update
Mask & Markup Concurrency	N/A	Strictly Forbidden	Strictly Forbidden
Known Weaknesses Targeted	Coarse edge blending	Faint background dot patterns	Text spatial alignment
Deployment Interfaces	API, ChatGPT	API, ChatGPT	Unreleased

Early community speculation highlighted by observers like Token Gremlin suggests that GPT Image 2.6 may also incorporate training updates targeting common complaints in 2.5, such as faint background dot patterns and complex non-Latin typography rendering.

Availability: How to Access GPT Image 2.6

Until OpenAI exposes an official model identifier, GPT Image 2.6 represents an active specification preview rather than a deployable production endpoint.

As of late October 2026, the official OpenAI Python SDK (version 3.41.0, published October 28) has not yet implemented the client bindings for the markup argument. Furthermore, the active production model list returned by OpenAI's API terminates at gpt-image-2.5-flare-2026-09-08. Developers cannot invoke the model in code until OpenAI publishes the corresponding runtime identifier.

# Current production SDK state
# Endpoint: POST /images/edits
# Status: GPT Image 2.6 identifier currently returns a model_not_found error

When OpenAI launches new generation checkpoints, rollout typically follows a predictable sequence:

1. Direct API Model Availability: The model ID (likely formatted along the lines of gpt-image-2.6 or a dated snapshot) becomes callable via the REST API.
2. Third-Party Platform Onboarding: Independent API platforms integrate the checkpoint for multi-model access. For instance, developers building multimodal pipelines can utilize GPT Image 2 via kie.ai's unified API today, with upcoming releases monitored for catalog integration.
3. ChatGPT Web and Mobile Integration: The underlying generation engine in ChatGPT's consumer interface is upgraded to the new model architecture.

What We Don't Know Yet

Because OpenAI has not accompanied the API specification update with an official announcement, several critical technical parameters remain unconfirmed:

Source: @Mr_Salio

• Arena Canary Identity: Between October 23 and October 25, 2026, community evaluators tracked anonymous model candidates on the LMSYS Arena platform, including shooting-star-red-alpha, shooting-star-green-alpha, and aurora-purple-alpha. Several commentators suggested these checkpoints represent canary evaluations for GPT Image 2.6, but OpenAI has not verified any connection.
• Pricing per Megapixel: Generation and edit pricing tiers remain unknown. GPT Image 2 and 2.5 operate under set resolution-based fees; it is unclear whether markup-assisted processing will carry a computational premium.
• Context Resolution Limits: The maximum resolution for both base images and markup overlays has not been confirmed, though current API endpoints enforce parity between input assets.
• Firm Launch Schedule: Speculative commentary across developer networks anticipates an imminent rollout, but OpenAI has provided no release roadmap.

Frequently Asked Questions

What is GPT Image 2.6?

GPT Image 2.6 is an upcoming image generation and editing model developed by OpenAI, identified in official API documentation through the introduction of a markup overlay editing parameter. It succeeds GPT Image 2.5 and provides dedicated capabilities for processing transparent sketch layers alongside text prompts.

When will GPT Image 2.6 be released?

OpenAI has not announced an official release date for GPT Image 2.6, though its appearance in the official API specification indicates that developer rollout is in active preparation. The model identifier is not yet active in production API endpoints.

How does markup-guided editing work in GPT Image 2.6?

Markup-guided editing allows developers to supply a transparent PNG overlay containing rough sketches, arrows, or color blocks that guide regional image transformations without requiring a standard binary inpainting mask. The model interprets the spatial positions and colors of the markings to constrain the generative output.

Can you combine the markup parameter with a mask in the OpenAI API?

No, OpenAI's API specification explicitly states that the markup parameter cannot be combined with a mask in a single image edit request. The API enforces strict mutual exclusivity between standard alpha masking and markup overlay inputs.

Is GPT Image 2.6 open source?

GPT Image 2.6 is a proprietary, closed-weights model accessible exclusively through OpenAI's hosted API endpoints and official consumer interfaces. Weights will not be released for local execution or self-hosted deployment.

How much does GPT Image 2.6 cost?

Pricing for GPT Image 2.6 has not yet been confirmed by OpenAI, as the model lacks an active public pricing schedule and callable model identifier. Full pricing metrics will be published once the model reaches general availability.

What happened to GPT Image 1?

OpenAI officially shut down the legacy gpt-image-1 API endpoint on October 23, 2026. All developer workloads were migrated to second-generation models, including GPT Image 2, GPT Image 2.5, and the upcoming GPT Image 2.6.

What to watch next: Track updates to the openai-openapi GitHub repository for the appearance of an active gpt-image-2.6 model identifier, monitor the release notes of the official OpenAI Python SDK beyond version 3.41.0 for client-side markup parameter support, and watch LMSYS Arena model de-anonymizations to see if the shooting-star test models correlate with this release.

Try related models on Kie.ai: GPT Image 2.5, Seedream 5.0 Pro and Flux 3.