Daybreak Blue

Reporting on OpenAI's Daybreak expansion states that Daybreak Blue is built on GPT-5.6 Sol and provides approved defensive-security access with safeguards tailored to that use case. A second tier, Daybreak Red, exposes the new GPT-5.6-Cyber model for vulnerability research, exploit development and red-team work, and Daybreak Blue is positioned as the recommended starting point for most security teams. Daybreak enrollment requires identity verification, account-security and monitoring controls, and compliance checks tied to approved use cases. Enrolled customers can also reach Daybreak Blue and Daybreak Red through Amazon Bedrock, and OpenAI claims GPT-5.6-Cyber set the highest success rate it has measured on the ExploitGym benchmark while completing 95% of advanced cybersecurity tasks in its evaluation.

CNBC reports that OpenAI expanded its Daybreak cybersecurity initiative into two access tiers, Daybreak Blue for defensive security work and Daybreak Red for more advanced operations, alongside a new GPT-5.6-Cyber model. The article frames Daybreak as a May launch that was extended after rivals including Anthropic's Project Glasswing, with recent AI-cybersecurity disclosures cited as motivation for tighter tiering. Daybreak Blue is described as giving vetted users access to OpenAI's advanced general-purpose models with safeguards altered to permit defensive security tasks. OpenAI quoted in the piece says it is putting frontier intelligence in the hands of trusted defenders before attackers can deploy offensive AI at scale, positioning Blue as the broader, lower-friction tier.

Vals AI's model directory lists Daybreak Blue with an August 10, 2026 release date, developer OpenAI, a roughly one-million-token context window and 128k max output tokens. Pricing is recorded at $5.00 per million input tokens and $30.00 per million output tokens, with text and image input supported, file input supported, and video input unsupported. On Vals' CyberBench v1.1 the model scores 72.98% with a ±5.26 margin, ranking ninth out of 42 proprietary models. The page reports a 11-minute-50-second latency figure for the evaluation run and notes the default provider as OpenAI, giving an independent third-party data point on the alias's capabilities and cost.

OpenAI's developer documentation describes Daybreak Blue as an alias for its flagship general-purpose models, calibrated with safeguards for defensive cybersecurity work. The page lists a 1,050,000-token context window, 128,000 max output tokens, a February 16, 2026 knowledge cutoff, and reasoning-token support. Input modalities are text and image; audio and video are unsupported, and access requires separate approval through the Daybreak program. Supported endpoints include Chat Completions, Responses, Realtime and Batch, with streaming, function calling and structured output features enabled. The documentation directs developers to a Daybreak enrollment page and to dedicated pricing details, positioning Daybreak Blue as a restricted, cybersecurity-focused configuration of OpenAI's frontier general-purpose lineup rather than a standalone base model.

BenchLM's ACCR Daybreak Blue leaderboard explicitly states that Daybreak Blue is an access configuration for GPT‑5.6 Sol, not a separate model, confirming the tier framing. The benchmark measures the share of approved advanced-cyber requests completed rather than refused under Daybreak Blue safeguards. The page is marked display-only and compiled from provider self-reports, so it is illustrative rather than a comparative ranking. The leaderboard lists GPT‑5.6 Sol at 2.0% on ACCR Daybreak Blue, matching the figure OpenAI published for the tier, and shows GPT‑6 Astra at 3.5% as the current top entry. BenchLM does not use these results to rank overall model cyber performance and notes only two models have been evaluated. The page helps disambiguate Daybreak Blue as a configuration rather than an independent model release.

OpenAI announced Daybreak Blue on August 10, 2026, defining it as an access tier that unlocks frontier general-purpose models, including GPT‑5.6 Sol, with safeguards tailored for authorized defensive security work. The tier is positioned as the recommended starting point for most defenders. Supported use cases include vulnerability discovery, secure code review, malware analysis, incident response, and patch validation. Daybreak Blue removes system-level guardrails so GPT‑5.6 Sol can engage with these tasks. Even without those guardrails, the model still refuses highly dual-use prompts such as pentesting production systems. OpenAI also shared an internal metric, the Advanced Cybersecurity Completion Rate, showing GPT‑5.6 Sol completes 2.0% of approved advanced-cyber requests under Daybreak Blue, versus 1.5% with default safeguards active. The company says GPT‑5.6 Sol delivers state-of-the-art cybersecurity performance, with Daybreak Blue designed to give approved defenders more of that capability in real-world security work. The tier operates alongside a separate Daybreak Red program that offers the purpose-built GPT‑5.6‑Cyber model.