GPT-Image-2.6 Sunburst and Flare: Setup and Complexity Guide

OpenAI launched GPT-Image-2.6 Sunburst and Flare with adjustable complexity, SVG and GIF outputs, and markup-guided editing.

By ailog · Published Nov 8, 2026 · 5 min read · How we work
In short
  • OpenAI introduced gpt-image-2.6-sunburst and gpt-image-2.6-flare on 2026-11-04 across generations and edits endpoints.
  • Both 2.6 releases add an adjustable complexity parameter, SVG generation, and animated GIF creation.
  • Markup-guided editing is now supported for compatible GPT Image 2.5 and 2.6 variants.
  • Text input costs $5 per 1M tokens ($1.25 cached), while image tokens run $8 input ($2 cached) and $30 output per 1M tokens.

OpenAI 2026-11-04 Image Generation Release Overview

On 2026-11-04, OpenAI introduced two additions to its image family in the API changelog: GPT-Image-2.6 Sunburst and GPT-Image-2.6 Flare. These models expand on the prior 2.5 series by bringing four core capabilities: an adjustable complexity configuration, vector SVG creation, animated GIF generation, and markup-guided image modifications.

Both models serve specific production roles. The Sunburst variant (gpt-image-2.6-sunburst) is designed for visual work requiring intricate composition and fine control over details. The Flare variant (gpt-image-2.6-flare) prioritizes fast iteration cycles for interactive visual tools. These releases operate across two routes in the REST interface: the image generation endpoint at v1/images/generations and the image modification endpoint at v1/images/edits.

Beyond new model IDs, OpenAI introduced markup-guided editing support across compatible GPT Image 2.5 and 2.6 releases. Teams building multi-step visual workflows or standalone asset generators can reference image generation APIs compared for architectural context across providers.

Key Capabilities: Complexity, Formats, and Markup Edits

The release introduces direct control over visual planning via the complexity request field. Rather than relying solely on prompt engineering to balance rendering fidelity and generation time, callers can specify how much layout planning the system performs before drawing pixels. The parameter accepts three string values: low, medium, and high.

  • Low complexity: Limits initial planning overhead. Ideal for targeted, isolated modifications like replacing background colors or updating text on an existing sign.
  • Medium complexity: Delivers balanced visual structure, preserving object shapes while processing secondary layout elements.
  • High complexity: Applies extensive composition planning. Recommended for multi-subject scenes, dense lettering, complex spatial layouts, and intricate physical interactions between items.
  • Default behavior: Omitting the complexity field defaults the generation to high.
  • Pricing impact: Modifying the complexity setting incurs no added API fee beyond standard token consumption.

Developers should note that complexity is distinct from the existing quality setting, which dictates rendering finish rather than planning depth. Higher complexity can extend processing latency and may introduce extraneous structural shifts if applied to simple edits, making evaluation necessary before enforcing defaults in production pipelines.

In addition to raster outputs, GPT Image 2.6 models support generating scalable vector graphics (SVG) and multi-frame animated GIFs directly from generation endpoints. Markup-guided editing also allows developers to supply structural markup to direct modifications precisely across both 2.5 and 2.6 model families.

Pricing Structure and Endpoint Availability

Unlike older legacy image endpoints that billed strictly on fixed per-image dimensions, both gpt-image-2.6-sunburst and gpt-image-2.6-flare share a unified token-based billing structure. Token metrics are divided into textual tokens and visual tokens, incorporating distinct rates for prompt caching.

Metric Input Type Price (per 1M tokens)
Text tokens Input $5
Text tokens Cached input $1.25
Image tokens Input $8
Image tokens Cached input $2
Image tokens Output $30

Developers monitoring operational spend across multimodal pipelines can model these costs against text LLM baselines using the /tools/llm-api-cost-calculator/ or consult the /tools/llm-api-pricing/ index. When running batch or conversational workflows, remember that GPT Image 2.6 snapshot endpoints are dedicated strictly to v1/images/generations and v1/images/edits.

According to the model documentation, routes such as v1/chat/completions, v1/responses, v1/batch, v1/fine-tuning, v1/realtime, and v1/embeddings are explicitly not supported directly by these model IDs. Rate limits follow account usage tiers listed in organization settings.

Implementation and Safety Identifier Migration

Integrating the 2.6 models into existing Image API pipelines requires updating the target model string to gpt-image-2.6-sunburst or gpt-image-2.6-flare in HTTP or SDK payloads. If you manage user safety, you must also adapt parameter naming conventions.

OpenAI deprecated the legacy user request parameter on image endpoints. Applications serving end users should instead transmit a stable, pseudonymized safety_identifier. This string should not contain cleartext personal data; hashing internal user identifiers or email addresses is recommended, with a hard length cap of 128 characters or fewer. A non-empty safety_identifier overrides the user field when both are present, though legacy requests with user remain supported.

curl -X POST "https://api.openai.com/v1/images/generations" \
    -H "Authorization: Bearer $OPENAI_API_KEY" \
    -H "Content-Type: application/json" \
    -d '{
        "model": "gpt-image-2.6-sunburst",
        "prompt": "A clean blueprint diagram of an mechanical watch movement",
        "complexity": "high",
        "safety_identifier": "usr_hashed_7f83b1657ff1fc53b92dc18148a1d65d"
    }'

Additionally, organizations must ensure they have completed API Organization Verification in their developer console settings before dispatching traffic to GPT Image models.

Comparing Sunburst vs Flare for Production Pipelines

Selecting between Sunburst and Flare depends on workflow latency budgets and compositional demands. While token pricing between the two versions is identical, their rendering dynamics serve distinct technical architectures.

Model ID Target Workload Supported Routes Complexity Support
gpt-image-2.6-sunburst Detailed visual work, complex compositions, precision editing v1/images/generations, v1/images/edits low, medium, high (default high)
gpt-image-2.6-flare Fast visual iteration, interactive editing, rapid mockups v1/images/generations, v1/images/edits low, medium, high (default high)

When building automated batch workflows or content pipelines (such as discussed in automated blog pipeline architecture), pairing low complexity with gpt-image-2.6-flare offers faster generation cycles. Conversely, publishing assets requiring precise typographic layouts or multi-character arrangement benefit from setting high complexity on gpt-image-2.6-sunburst.

Migration and Deployment Checklist

Before switching production traffic to the GPT-Image-2.6 series, review operational configurations to avoid runtime rejection and unintended token expenditures.

  • Verify organization access: Complete the API Organization Verification process via the developer console general settings.
  • Update model strings: Replace legacy model names with gpt-image-2.6-sunburst or gpt-image-2.6-flare in generation and edit calls.
  • Tune complexity levels: Benchmark low, medium, and high on representative prompts; set low for isolated edits to avoid unwanted structural drift.
  • Migrate user parameters: Transition all tracking fields from the deprecated user attribute to safety_identifier, ensuring strings are hashed and under 128 characters.
  • Validate endpoint paths: Verify calls route to v1/images/generations or v1/images/edits, avoiding unsupported interfaces like v1/chat/completions or v1/batch.
  • Monitor cached token usage: Review billing dashboards to evaluate how prompt caching discounts ($1.25 text cached, $2 image cached) affect overall unit economics.
Sources
  1. OpenAI API changelog — 2026-11-04
  2. GPT-Image-2.6 Sunburst
  3. GPT-Image-2.6 Flare
  4. Image generation

Related