OpenAI Images 2.6 Adds GIFs and Two New API Models

OpenAI released Images 2.6 on November 4, adding animated GIF output, SVG vector export, markup-guided editing, and two new API models: GPT-Image-2.6 Flare and GPT-Image-2.6 Sunburst. The update ships immediately in ChatGPT and Codex across all tiers, and OpenAI claims 50% faster generation — though no independent benchmark rankings exist yet. For developers, the real story is in what you can and cannot ship to production today.

Flare vs Sunburst: Which OpenAI Images 2.6 Model Do You Need?

Images 2.6 introduces a split that GPT Image 2.5 didn’t have: a speed tier and a quality tier. GPT-Image-2.6 Flare targets latency-sensitive pipelines, hitting roughly six seconds at the low quality setting — suitable for rapid product image variants or high-volume draft generation. GPT-Image-2.6 Sunburst prioritizes detail and edit consistency across multiple revision rounds, making it the right call for brand assets, character work, or any workflow where visual coherence across iterations matters.

Moreover, both models share identical pricing: $5 per million text input tokens, $8 per million image input tokens, and $30 per million image output tokens. Neither supports the Responses API, Chat Completions, or Batch API. The choice is straightforward: use Flare when throughput matters, Sunburst when fidelity does.

import openai
client = openai.OpenAI()

# Flare: fast drafts and high-volume pipelines
draft = client.images.generate(
    model="gpt-image-2.6-flare",
    prompt="Product on white background, studio lighting",
    quality="low"
)

# Sunburst: brand assets and multi-round editing
final = client.images.generate(
    model="gpt-image-2.6-sunburst",
    prompt="Brand mascot in office setting",
    quality="high"
)

New Output Formats: GIFs, SVGs, and Markup

Furthermore, three new output capabilities shipped alongside the new models. GIF generation produces animated loops from 2 to 64 frames, with each frame running 20 to 2,000 milliseconds (rounded to 10ms increments). The defaults — 9 frames at 200ms each — produce a 1.8-second loop. SVG export generates static vector graphics via the generation endpoint only; editing does not support SVG output, and no scripts or embedded raster images are allowed inside the SVG. Markup editing adds a third new capability: submit a transparent PNG annotation layer alongside your source image to give the model spatial guidance.

However, these constraints matter. The markup annotation caps at 16,777,216 pixels and 50MB, and cannot combine with masks or the prompt_rewriter parameter. SVG billing also double-dips: OpenAI charges the image model rate for the reference image and the authoring model rate for the vector conversion. Know these limits before designing a pipeline around them. For more context on how Images 2.6 rolled out globally, CryptoBriefing has the full availability breakdown.

Related: OpenAI Kills Agent Builder, Evals API, and Stored Prompts — Nov 30 Checklist

What Is Not Ready Yet for Developers

The GIF feature is the headline — just not on the API, not yet. According to OrcaRouter’s technical analysis, “GIF support is behind a feature flag that is currently disabled, and it has not yet been publicly documented in the API.” The underlying code exists; third-party apps simply cannot rely on it yet. Wait for official documentation before shipping anything that depends on GIF output.

Additionally, the more significant gap is the Batch API removal. GPT Image 2.5 Batch was priced at $15 per million output tokens — half the standard rate. Images 2.6 drops Batch API support entirely, reverting all requests to the full $30 per million. For teams running bulk image jobs, that is not a feature upgrade — it is a silent 2x cost increase. High-volume pipelines should stay on 2.5 until OpenAI restores batch pricing or announces a volume tier. Full technical specs are covered in kingy.ai’s detailed breakdown of Flare and Sunburst.

One additional gotcha: Microsoft shipped MAI-Image-2.6 in August 2026, creating a naming collision with OpenAI’s release. The API identifiers differ — gpt-image-2.6-flare and gpt-image-2.6-sunburst for OpenAI vs mai-image-2.6 for Microsoft — but documentation confusion and tooling mismatches are already surfacing. Double-check your model identifier before any deployment.

Key Takeaways

• Use Flare for speed-critical pipelines (6-second generation, high-volume drafts); use Sunburst for quality and edit consistency across revision rounds
• GIF output exists in the API codebase but remains behind an undocumented feature flag — do not ship it to production yet
• Images 2.6 drops Batch API support entirely: bulk workflows that relied on the $15/M output token rate now pay $30/M — stay on GPT Image 2.5 for batch jobs
• Watch for naming confusion with Microsoft MAI-Image-2.6; always use the gpt-image-2.6-flare or gpt-image-2.6-sunburst model identifiers to confirm you are calling OpenAI’s models
• No independent benchmark rankings (Artificial Analysis, Arena) exist for 2.6 yet — treat OpenAI’s 50% speed claim as self-reported until third-party data arrives