Current model · OpenAI
GPT-6 Cyber benchmarks, pricing and specs
GPT-6 Cyber is an OpenAI model released on November 5, 2026 that takes text and image input and produces text. It costs $20 per 1M input tokens and $100 per 1M output tokens and accepts up to 1,050,000 tokens of context.
- Per 1M input tokens
- $20Listed today
- Per 1M output tokens
- $100Listed today
- Context window
- 1.05M#1 of 67 current models
Data verified November 6, 2026 · Dataset 2026-11-06
Specs
Key facts
- Provider
- OpenAI
- Model family
- GPT Cyber
- Release date
- November 5, 2026
- Status
- Current
- Access
- limited access
- API model id
- gpt-6-cyber
- Context window
- 1,050,000 tokens
- Max output
- 128,000 tokens
- Input
- text, image
- Output
- text
- Task families
- text, multimodal
- Declared capabilities
- reasoning
Charts
Against every current text model
Where GPT-6 Cyber lands among the current models of its task, on every independent benchmark, on price and on context.
Where it stands
57 of 58 list a lower price per 1M input tokens. Each bar is the share of current models it beats.
Price and context
Price per 1M input tokens
#58of 58
$20 · cheaper than 0%
Price per 1M output tokens
#58of 58
$100 · cheaper than 0%
Context window
#1of 67
1.05M tokens · longer than 85%
Benchmarks
Independent benchmark results
Scores come from independent evaluators that ran the model themselves, pinned to a dated snapshot. Rank is among current models in the same snapshot. OpenAI's own launch figures are in the next section, because they are not measured the same way.
No independent benchmark has published a result for GPT-6 Cyber yet. It is still compared below on price, context and capabilities.
Provider results
Reported by OpenAI
OpenAI's own results for GPT-6 Cyber, beside the rivals it chose, as its table lists them. Labs pick the benchmarks and run them their own way, so compare within one table only. Internal benchmarks and other evaluators' indexes are left out.
GPT-6 Cyber system card, November 5, 2026OpenAI · 2026-11-06
| Benchmark | GPT-6 Cyber | GPT-6 Astra | GPT-5.6 Sol |
|---|---|---|---|
| ExploitBenchmean capability score, normalized to arbitrary code execution | 100.0% | 100.0% | 78.5% |
| SEC-Bench Propass@1 | 87.0% | 85.4% | 79.1% |
| SRE-Benchall six objectives, pass@1 | 92.8% | 88.0% | 55.9% |
OpenAI's system card results at the highest reasoning budget it tested. The GPT-6 Astra and GPT-5.6 Sol scores match the runs OpenAI's launch post labels Daybreak Blue, the access tier without their cyber guardrails. OpenAI warns that ExploitBench's historical vulnerabilities may have been seen in training. ExploitGym, where the system card gives the same scores, is in the launch post table. Left out here: evaluations OpenAI built itself (ExploitBench Internal Port, ExploitGym Hardened, ExpertSWE) and Irregular's private suites.
GPT-6 Cyber launch post, November 5, 2026OpenAI · 2026-11-06
| Benchmark | GPT-6 Cyber | GPT-5.6 Cyber | GPT-6 Astra | GPT-5.6 Sol |
|---|---|---|---|---|
| ExploitGymstrict supplied-vulnerability pass@1 | 48.5% | 32.1% | 42.4% | 30.3% |
OpenAI's chart from the GPT-6 Cyber announcement. GPT-6 Astra and GPT-5.6 Sol ran with Daybreak Blue access; with its standard safeguards GPT-6 Astra refuses these tasks and scores 0%. Left out here: RemoteExploitBench and the web pentesting evaluation, which OpenAI built itself, the honeypot evaluation, where lower is better, and SRE-Bench, which is in the system card table.
Fit
Strengths, cautions and best uses
- Strengths
- OpenAI's cybersecurity model for authorized vulnerability research and exploit development: a variant of GPT-6 Astra trained to find and validate exploitable vulnerabilities while staying within the scope a researcher sets
- OpenAI reports 48.5% on ExploitGym, against 32.1% for GPT-5.6 Cyber and 42.4% for GPT-6 Astra, and 92.8% on SRE-Bench reverse engineering against 88.0% for Astra
- OpenAI says it built working exploits for 948 Linux kernel vulnerabilities, more than five times the 166 GPT-5.6 Sol managed with the same harness, and found 81 previously unknown high-severity Firefox vulnerabilities, many already fixed by Mozilla
- 1,050,000-token context window and 128,000 max output tokens, at $20 input, $100 output and $2 cached input per 1M tokens, twice the standard rates of GPT-6 Astra
- Cautions
- Limited access: OpenAI is rolling it out through the API and Codex to select approved Daybreak Red customers first, with identity verification, single sign-on, multi-factor authentication and API IP allowlisting
- OpenAI rates it Critical in cybersecurity under its Preparedness Framework, like GPT-6 Astra and up from High for GPT-5.6 Cyber, and adds checks that block high-risk activity against critical systems
- It does not beat GPT-6 Astra at general coding (82.95% against 83.24% on OpenAI's internal ExpertSWE), and on Irregular's FrontierCyber it solved 90 of 288 challenges against 96 for Astra; no independent benchmark score is recorded here yet
- Best for
- Security teams approved for Daybreak Red that develop and validate exploits in authorized tests
- Vulnerability research and patch validation across large codebases such as browsers and kernels
From benchmark to production
Building on GPT-6 Cyber?
Public scores rarely hold on your own data. We test it against the alternatives on your real tasks, then build the agent or pipeline around the model that wins.
AI Model Cost AuditLaunch price $100, regular price $150What the audit includes
Alternatives
Alternatives to GPT-6 Cyber
The closest current text models by price. The cheapest model that scores as well or better is listed first; failing that, the cheapest one within 3% of the benchmark's scale.
On score and price
GPT-6 Cyber beside the 4 alternatives listed below, on Arena Text and on blended price per 1M tokens.
Arena Text
Measured October 28, 2026, higher is better
- GPT-6 Cyber
- Not scored
- GPT-Rosalind-6
- Not scored
- Claude Fable 5
- 1,504
- Claude Fable 5.1
- 1,504
- GPT-6 Astra
- 1,476
Blended price per 1M tokens
Listed today, lower is better
- GPT-6 Cyber
- $40
- GPT-Rosalind-6
- $30
- Claude Fable 5
- $20
- Claude Fable 5.1
- $20
- GPT-6 Astra
- $20
GPT-Rosalind-6
- 25% cheaper per 1M input tokens, $15 vs $20
Claude Fable 5
- 50% cheaper per 1M input tokens, $10 vs $20
- From Anthropic
Claude Fable 5.1
- 50% cheaper per 1M input tokens, $10 vs $20
- From Anthropic
GPT-6 Astra
- 50% cheaper per 1M input tokens, $10 vs $20
Further reading
More on GPT-6 Cyber
FAQ
GPT-6 Cyber FAQ
How much does GPT-6 Cyber cost?
GPT-6 Cyber costs $20 per 1M input tokens and $100 per 1M output tokens at the standard tier. It is the most expensive of 58 current text models priced per 1M input tokens. Source: OpenAI, verified November 6, 2026.
What is the context window of GPT-6 Cyber?
GPT-6 Cyber accepts up to 1,050,000 tokens of context. It can return up to 128,000 tokens in one response. That ranks #1 of 67 current text models by context length.
What benchmark results has OpenAI published for GPT-6 Cyber?
In the GPT-6 Cyber system card (November 5, 2026), OpenAI reports 100.0% on ExploitBench (mean capability score, normalized to arbitrary code execution), 87.0% on SEC-Bench Pro (pass@1) and 92.8% on SRE-Bench (all six objectives, pass@1). These are OpenAI's own figures, beside the rivals it chose, not independent measurements.
Is GPT-6 Cyber open source or API only?
Neither. OpenAI offers GPT-6 Cyber only to organizations it approves, with no public API and no published weights; the cautions on this page say who can apply.
What are the best alternatives to GPT-6 Cyber?
The closest current alternatives by price and measured performance are GPT-Rosalind-6 (25% cheaper per 1M input tokens, $15 vs $20), Claude Fable 5 (50% cheaper per 1M input tokens, $10 vs $20), Claude Fable 5.1 (50% cheaper per 1M input tokens, $10 vs $20) and GPT-6 Astra (50% cheaper per 1M input tokens, $10 vs $20).
Sources
Sources
- GPT-6 Cyber: Advancing the frontier of aligned cybersecurity agents · OpenAI · official · verified November 6, 2026
- GPT-6 Cyber System Card · OpenAI · official · verified November 6, 2026
- GPT-6-Cyber model · OpenAI · official · verified November 6, 2026
All 255 sources behind the benchmarks, with their versions, licenses and notes, are in the source catalog.
Cite this page
Van Data Team. "GPT-6 Cyber Benchmarks, Pricing and Specs." AI Model Benchmark Dataset 2026-11-06. https://vandatateam.com/ai-model-benchmarks/gpt-6-cyber
Reuse the data: JSON · CSV · or explore every model in the AI model benchmark comparison.
Production evaluation
Deciding whether GPT-6 Cyber fits your workload?
Public benchmarks narrow the shortlist. We test the finalists on your own tasks, data and cost limits, then build the routing and review controls to run the winner in production.
- Evaluation cases and acceptance thresholds tied to real work
- Model routing, fallbacks and cost controls