GPT-5.5-Cyber and Patch the Planet bring frontier-model capability to defensive security work
If you write or review code for a living, this is the OpenAI release that touches your day job directly. GPT-5.5-Cyber is a security-tuned version of GPT-5.5 that sets a new high on the CyberGym benchmark at 85.6 per cent, up from 81.8 per cent for the standard model. It ships alongside a Codex Security plugin that bakes vulnerability scanning straight into the editor where you already write code.
Access is gated behind OpenAI's Trusted Access for Cyber programme, which loosens the usual automated refusals for approved defensive work including code review, vulnerability triage and penetration testing, while still blocking offensive tasks like credential theft and malware deployment. In a live demonstration, Trail of Bits used the model to stand up a full fuzzing lab in under a day and surfaced two dozen local privilege escalation exploits in the Linux kernel. The strategic read is that OpenAI now wants to own both sides of the developer workflow: the writing and the securing.
Alongside the model, Patch the Planet is an open-source initiative run with Trail of Bits and HackerOne that has already signed up more than 30 projects including cURL, Go, Python and Sigstore. A five-day opening sprint merged dozens of patches across critical infrastructure. It is a direct answer to Anthropic's Project Glasswing, and the two firms look set to keep racing on security tooling.
Fable 5's free window closes: metered pricing at $10 per million input tokens begins from 23 June
If you have been trialling Fable 5 inside a paid Claude plan, the meter started running yesterday. From 23 June the model moves to usage credits at $10 per million input tokens and $50 per million output tokens, after the free window bundled into Pro, Max, Team and Enterprise seats closed the day before. That pricing lands at roughly double Opus 4.8, so the practical question is no longer whether Fable 5 is impressive but whether its output justifies the premium over routing simpler jobs to cheaper models.
Anthropic did not extend the free window to compensate for the six days the model was offline during the mid-June suspension under the US export-control directive, so team budgets need recalculating now. For anyone costing out an AI roadmap, Fable 5 is now a real line item rather than a free upgrade, and the natural next step for most teams is internal benchmarking to identify exactly which task types earn the premium back.
Claude suffered a roughly three-hour outage on 23 June, with elevated error rates across models from around 15:02 BST. Anthropic confirmed it was resolved by 17:44 BST. Service is back to normal; flagged only because it briefly affected access across all Claude tiers. TechRadar, 23 Jun →
SpaceX signed a compute deal worth up to $6.3 billion with open-source startup Reflection AI on 23 June, at $150 million a month for GB300 access at its Colossus 2 site. It is relevant to anyone watching Grok because it sharpens the open question of whether SpaceX is prioritising selling compute over backing its own frontier model. AIToolsRecap, 23 Jun →
Today's themes
Cybersecurity has become the headline battleground for frontier labs. OpenAI's GPT-5.5-Cyber and Patch the Planet are an explicit response to Anthropic's Glasswing, and both firms are betting that fixing real vulnerabilities in critical open-source infrastructure is the most useful and most defensible application of frontier capability. The second theme is monetisation discipline: Anthropic ending Fable 5's free window signals that the era of generous trial access to top-tier models is tightening. Builders should expect to pay premium rates for premium output, and internal benchmarking to understand exactly where the premium is justified is no longer optional. Finally, the SpaceX and Reflection AI compute deal is a reminder that the economics underneath these products are shifting fast, with compute capacity itself becoming a tradeable, revenue-generating asset that could reshape which labs can afford to keep training at the frontier.