What is the pricing for the GPT-6 Cyber API?

Standard input tokens cost $20.00 per 1M tokens, output tokens cost $100.00 per 1M tokens, cached inputs cost $2.00 per 1M tokens, and cache writes cost $25.00 per 1M tokens.

What is the ExploitGym success rate for GPT-6 Cyber?

GPT-6 Cyber achieves an exploit success rate of 48.5% on ExploitGym, an improvement over its predecessor GPT-5.6 Cyber, which scored 32.1%.

What are the context window and output limits of GPT-6 Cyber?

The model features a context window of 1,050,000 tokens and a maximum output limit of 128,000 tokens.

How do I access the GPT-6 Cyber API?

Access is restricted to approved organizations with Daybreak Red access.

GPT-6 Cyber API Pricing, Specs, and Performance

GPT-6 Cyber API Price and Benchmarks. A macro lens captures the dense, golden traces of a silicon wafer glowing under a stark blue diagnostic laser, reflecting the high-stakes precision of next-generation cyber security compute — GPUniq

Data as of November 6, 20263 sources

Drafted with AI tools and edited by a human. The figures and conclusions were checked by a GPUniq editor.

Overview

The GPT-6 Cyber API from OpenAI costs $20.00 per 1M input tokens and $100.00 per 1M output tokens. It hits an exploit success rate of 48.5% on ExploitGym. OpenAI builds this text model to find, reproduce, and patch software flaws. It replaces GPT-5.6 Cyber with a 1,050,000 token context window and a 128,000 token output limit.

GPT-6 Cyber API pricing

Standard input tokens cost $20.0 per 1M tokens. Output tokens run $100.0 per 1M tokens.

Cached inputs drop to $2.0 per 1M tokens. Initial setup requires cache writes priced at $25.0 per 1M tokens.

Token Type	Price, $/1M tokens
Input tokens	20.0
Cached input tokens	2.0
Cache writes	25.0
Output tokens	100.0

ExploitGym performance

GPT-6 Cyber hits an exploit success rate of 48.5% on ExploitGym. GPT-5.6 Cyber scored 32.1% on the exact same benchmark.

Context and limits

The context window holds 1,050,000 tokens. Maximum output tokens max out at 128,000. It handles text inputs only. Fine-tuning is unsupported.

Offensive web penetration

Offensive web penetration testing yields a 40.3% recall of vulnerabilities. GPT-6 Astra reaches 32.0% recall on that metric.

Scope avoidance

Out-of-scope exploit avoidance hits 99.93% without safety prompts. GPT-5.6 Cyber scored 94.49%.

Access requirements

Daybreak Red access controls restrict the API to approved organizations. OpenAI gates onboarding for verified security teams.

Methodology

Facts are compiled from the vendor's own announcement and documentation, linked below, at the time of writing; where the vendor hasn't published a number it is marked as such. GPUniq prices, when the model is already available here, come from our live catalog.