OpenAI
GPT-6 Cyber
OpenAI announced GPT-6 Cyber on 2026-11-05. The API id is gpt-6-cyber, at $20 per million input tokens and $100 per million output tokens, and calling it requires Daybreak Red access plus approval. No independent tracked score is stored as of 2026-11-06. OpenAI's ExploitGym result of 48.5% is its own claim, and the post does not retire GPT-6 Astra.
GPT-6 Cyber benchmarks and pricing, every number sourced: no GPT-6 Cyber benchmark scores tracked yet, priced at $20.00 per million input tokens and $100.00 per million output.
- Released
- 2026-11-05
- Availability
- API id gpt-6-cyber; Daybreak Red access and approval required. Not a general public launch (checked 2026-11-06).
- License
- proprietary
- Context window
- 1M tokensvendor-stated (developers.openai.com/api/docs/models/gpt-6-cyber: 1,050,000-token context, 922,000 max input, 128,000 max output; checked 2026-11-06)
- Knowledge cutoff
- Not disclosed
- Parameters
- Not disclosed
- Architecture
- Not disclosed
GPT-6 Cyber announced pricing
$20.00 in / $100.00 out per 1M tokens — official pricing source
What GPT-6 Cyber costs per job
| Workload | Tokens in / out | Cost |
|---|---|---|
| One long chat turn | 100K / 10K | $3.00 |
| A codebase review | 1,000K / 100K | $30.00 |
| A day of agent work | 10,000K / 1,000K | $300.00 |
Computed from GPT-6 Cyber’s list rates above — cache discounts and batch tiers are not applied.
GPT-6 Cyber is one of 8 OpenAI models tracked on this site, at these official list prices.
| Model | In / 1M | Out / 1M |
|---|---|---|
| GPT-6 Cyber | $20.00 | $100.00 |
| GPT-6.1 Terra | $1.00 | $5.00 |
| GPT-6.1 Sol | $2.00 | $10.00 |
| GPT-6 Luna | $0.10 | $0.50 |
| GPT-6 Sol(previous version) | $2.00 | $10.00 |
| GPT-6 Astra | $10.00 | $50.00 |
| GPT-5.6 Luna(previous version) | $0.20 | $1.20 |
| GPT-5.6 Sol(previous version) | $4.00 | $20.00 |
Notes on the record
OpenAI announced GPT-6 Cyber on 2026-11-05 and calls it a cyber-specialized variant of GPT-6 Astra for authorized security work, not a replacement. The public API id is gpt-6-cyber. Calling it requires Daybreak Red access and approval. First access is the API and Codex for a select group of approved Daybreak Red customers, later expanding to all Daybreak Red customers. GPT-5.6 Cyber is being opened to all Daybreak Red customers too. The post does not retire GPT-6 Astra, GPT-6 Sol, GPT-6 Luna, or either GPT-6.1 model. Checked 2026-11-06.
The stored price is $20 per million input tokens and $100 per million output tokens, which OpenAI calls the standard short-context rate. Cached input is $2 per million. Cache writes are $25 per million for 30 minutes or $50 per million for 12 hours. Every long-context cell for gpt-6-cyber on the pricing page is blank. That blank is not a second rate, and it does not show the short-context rate holding past 272,000 input tokens, OpenAI's own boundary. Batch is not supported.
The model page states a 1,050,000-token context window, maximum input 922,000, and maximum output 128,000. Input is text and image; output is text. Reasoning tokens, streaming, structured outputs, function calling, web search, and prompt caching are listed. No knowledge cutoff is stated on the GPT-6 Cyber model page.
OpenAI classifies GPT-6 Cyber Critical in cybersecurity, the same designation as GPT-6 Astra and up from High for GPT-5.6 Cyber. ExploitGym intended-exploit success is 48.5% against 32.1% for GPT-5.6 Cyber on the launch post, and 48.45% against 42.43% for GPT-6 Astra on the system card — two baselines, not one comparison. The honeypot stays inside scope in 99.93% of graded rollouts without a deployment prompt and in 100% with one; the system card's matching exploitation rates are 0.07% and 0%. These are vendor self-reports on evals this site does not track, so none is stored as a score.
Compare with
FAQ
Who can call GPT-6 Cyber?
Not the general public, as of 2026-11-06. The API id gpt-6-cyber is documented, but the model page says the organization needs Daybreak Red access and approval. The launch post says the first customers are a select approved group, on the API and in Codex, with access planned to widen to all Daybreak Red customers. Rate-limit rows on the docs page apply to those approved organizations. They are not a public free tier.
What does GPT-6 Cyber cost?
The published text rate is $20 per million input tokens and $100 per million output tokens, which OpenAI labels a short-context price. Cached input is $2 per million. Cache writes are $25 per million for a 30-minute window or $50 per million for 12 hours. The Cyber models table leaves the long-context columns blank, so this page does not invent a price past 272,000 input tokens and does not treat the short-context rate as valid at the 922,000-token maximum input. Batch is not supported.
What is GPT-6 Cyber's context window?
1,050,000 tokens, stated on the model page, with a 922,000-token maximum input and a 128,000-token maximum output. Input modalities are text and image, and output is text. No knowledge cutoff is disclosed on that page. The window matches the other current GPT-6 rows this site lists, but the long-context price for GPT-6 Cyber is unpublished.
Has GPT-6 Cyber been independently benchmarked?
No tracked-benchmark row exists for GPT-6 Cyber as of 2026-11-06. OpenAI's own ExploitGym numbers are 48.5% versus 32.1% for GPT-5.6 Cyber on the launch post, and 48.45% versus 42.43% for GPT-6 Astra on the system card. Its honeypot claim is 99.93% inside scope without a deployment prompt and 100% with one. Those figures stay in prose. They are not independent runs, and ExploitGym is not a benchmark this site tracks.
Does GPT-6 Cyber replace GPT-6 Astra?
No. The system card calls GPT-6 Cyber a cyber-specialized variant of GPT-6 Astra, and neither that card nor the launch post says Astra, Sol, Luna, or the GPT-6.1 models are retired. This site does not mark any of them a previous version because of this release. Astra stays a mainline model; the launch post still points at it for code review and patching.
What is GPT-6 Cyber for?
OpenAI positions GPT-6 Cyber for authorized cybersecurity work. The changelog says to investigate vulnerabilities, reproduce them in an authorized test environment, and verify fixes. The system card names vulnerability discovery, exploit development, and web pentesting, with extra post-training meant to reduce cyber refusals while staying inside the user's authorized scope. That is the vendor's job description. This site has no independent measurement of that job.
Further reading
- Models with 10M token context windows 2026 — GPT-6 Cyber is one of the 47 models it compares.
- Self-reported vs. independent AI benchmarks — the self-reported-versus-independent rule every score on GPT-6 Cyber’s page follows.