GPT Image 2.6 & GPT-6 Cyber: API Pricing and Access
OpenAI lists GPT Image 2.6 Flare, Sunburst and GPT-6 Cyber API rates. Compare token costs, endpoint support and approval requirements.
By AI Pricing Guru Editorial Team
The AI Pricing Guru editorial team verifies prices against official provider pages and maintains the datasets, calculators and guides. See our methodology and contact us to report a correction.
Report summary — 2026-11-10
- GPT Image 2.6 Flare and Sunburst have identical published Standard token rates; choose through workload testing, not an assumed price difference.
- GPT-6 Cyber requires Daybreak Red access and organization approval; a published rate card does not mean your account can call it.
- The image models support image generation and editing, but their current docs mark Batch and Responses unsupported.
GPT Image 2.6 versus 2.5: Standard token rates
| Model | Text input | Cached text | Image input | Cached image | Image output |
|---|---|---|---|---|---|
| GPT-Image-2.6 Flare Price: Source verified: 2026-11-12 01:01:11 UTC. Official source | $5.00 | $1.25 | $8.00 | $2.00 | $30.00 |
| GPT-Image-2.6 Sunburst Price: Source verified: 2026-11-12 01:01:11 UTC. Official source | $5.00 | $1.25 | $8.00 | $2.00 | $30.00 |
| GPT-Image-2.5 Flare Price: Source verified: 2026-11-12 01:01:11 UTC. Official source | $5.00 | $1.25 | $8.00 | $2.00 | $30.00 |
| GPT-Image-2.5 Sunburst Price: Source verified: 2026-11-12 01:01:11 UTC. Official source | $5.00 | $1.25 | $8.00 | $2.00 | $30.00 |
USD per million tokens of each modality, not per generated image. Cached-input rates require an eligible request. Rates come from the canonical pricing API and the official rate card.
Cost comparison from recorded API rates
USD per 1M tokens. Input and output rates are charted separately; inclusion is not an availability or subscription-access guarantee.
Cyber models: Standard short-context token rates
| Model | Provider | Input / 1M | Cached / 1M | Output / 1M |
|---|---|---|---|---|
| GPT-6 Cyber Price: Last source verification (overdue): 2026-11-09 22:38:41 UTC. Official source | openai | $20.00 | $2.00 | $100.00 |
| GPT-5.6 Cyber Price: Last source verification (overdue): 2026-11-09 20:37:42 UTC. Official source | openai | $12.50 | $1.25 | $75.00 |
Built from pricing.json at publish time.
Dated story text, scoped to the report date above:
What changed
OpenAI’s current rate card and per-model documentation now list GPT Image 2.6 Flare, GPT Image 2.6 Sunburst and GPT-6 Cyber. These are API listings, not evidence of access through a ChatGPT subscription. The sources do not establish an original launch date.
The maintained tables above show the new models beside their predecessors. Both Image 2.6 variants retain the listed Image 2.5 Standard token rates. GPT-6 Cyber has higher fresh-input, cached-input and output rates than GPT-5.6 Cyber. A new listing alone does not retire the older API models.
Image 2.6: same rate card, different workflow emphasis
OpenAI describes Flare as image generation and editing for fast iteration, and Sunburst as image generation and editing for detailed visual work. Both accept text and images and produce images. Those descriptions are useful evaluation starting points, not independent evidence of a quality or latency advantage.
Use gpt-image-2.6-flare or gpt-image-2.6-sunburst with the documented v1/images/generations and v1/images/edits endpoints. Their current endpoint tables explicitly mark Batch and Responses as unsupported. Do not carry a predecessor’s endpoint support or Batch discount into an Image 2.6 budget.
The table separates text-input, cached-text, image-input, cached-image and image-output tokens. It is not a fixed price per image. Actual cost depends on the tokens used by the request and output; a cached-input rate is not a promise every repeated input qualifies. Use OpenAI’s image-generation guide for request-specific estimates, then compare measured usage on representative edits and generations.
GPT-6 Cyber: access before budgeting
GPT-6 Cyber is documented for vulnerability investigation, reproduction in authorized test environments and fix verification. Your organization needs Daybreak Red access and approval; OpenAI says your account team can confirm eligibility. Set up the organization and project before attempting production integration.
The model supports Chat Completions and Responses. Its documentation marks Batch unsupported. The published input, cached-input and output figures in the table are Standard short-context rates, not a blanket quote for every context length or service configuration.
Cache reads and cache writes are different billing categories. The canonical GPT-6 Cyber model page lists the maintained 30-minute write rate. OpenAI also publishes a separately priced 12-hour write option in the official Cyber documentation. Do not use the cheaper cache-read rate as the cost of creating a cache, or assume the two retention options cost the same.
What developers should do now
Image teams: test Flare and Sunburst on the same prompts, edits and acceptance criteria. Since the Standard rate cards match, compare total tokens, retries, latency and accepted outputs before choosing. Existing Batch or Responses workflows need an endpoint compatibility check before migration.
Security teams: confirm Cyber approval first. Then evaluate against your current model on authorized tasks, counting output length, cache writes, tool charges and reviewer time. Higher token rates need a measured improvement in useful results to justify migration.
Existing users: keep validated routes until replacement behavior is tested. This update does not establish a shutdown deadline or subscription entitlement. Use the OpenAI pricing page for the complete catalog and Anthropic pricing for an external budget baseline. The token calculator can estimate supported text-token workloads; it is not a fixed-price Image 2.6 estimator. Our API pricing comparison guide explains broader provider tradeoffs.
Sources and verification.
Official sources: API pricing, Flare model docs, Sunburst model docs and Cyber model docs. The source evidence for this report was checked November 9, 2026; the tables read maintained canonical data at build time and display their recorded verification dates. Publication time is not a fresh source check.