Two announcements. One week. Very different implications depending on who you are.
OpenAI announced GPT-5.5 on April 23, 2026, and simultaneously launched Workspace Agents for enterprise customers around April 22. These aren’t one story, they’re three: a consumer reasoning upgrade, an enterprise automation platform, and a safety classification that compliance teams need to read carefully.
Start with the safety classification, because it’s the most concrete and most verifiable development in this release. OpenAI’s deployment safety documentation states the company is “treating the biological/chemical and cybersecurity capabilities of GPT-5.5 as High under our Preparedness Framework”, and explicitly notes GPT-5.5 sits below the Critical threshold for cybersecurity. That distinction matters. “High” under the Preparedness Framework means enhanced monitoring, access controls, and use-case restrictions are in place. “Critical” would mean the model doesn’t ship. GPT-5.5 shipped.
GPT-5.5 Thinking is available to ChatGPT Plus, Pro, Business, and Enterprise users. OpenAI reports the model matches GPT-5.4 in per-token latency while using fewer tokens on complex reasoning tasks, though this claim traces to OpenAI’s own statements and the primary announcement page was temporarily inaccessible at time of publication. API access is described as forthcoming, with pricing reportedly at $5 per million input tokens and $30 per million output tokens. A new $100 per month subscription tier for heavy Codex and Pro use was reportedly introduced alongside the release. Neither the API pricing nor the new subscription tier has been independently confirmed.
GPT-5.5’s context window generated conflicting reports, one third-party source cited a 1 million token capacity while the Wire’s structured data stated 256,000 tokens for the Pro tier. The specific figure for each tier wasn’t confirmed at time of publication and isn’t included here.
Workspace Agents are the enterprise story. According to aintelligencehub.com and corroborated by VentureBeat, the launch went live around April 22, enabling automation across Slack, Salesforce, and Gmail. Workspace Agents are described as including human-in-the-loop approval steps for enterprise workflows, the precise mechanism wasn’t directly quoted in available reporting, but the enterprise access controls and admin oversight framing appeared consistently across sources.
Why does this matter now? The Preparedness Framework classification is the signal practitioners should act on first. OpenAI’s voluntary safety architecture is one of the few public frameworks describing how a frontier lab internally gates dangerous capability deployment. “High but below Critical” in cybersecurity means GPT-5.5 can reason about vulnerabilities in ways that prompted internal concern, enough to flag, not enough to withhold. That’s a meaningful data point for security teams evaluating whether GPT-5.5 belongs in their environments, and for compliance professionals assessing agentic AI under the EU AI Act.
For enterprise IT teams, Workspace Agents represent OpenAI’s most direct move yet into workflow automation territory occupied by Microsoft Copilot and Salesforce Einstein. The Slack and Gmail integrations aren’t incidental, they’re the deployment surface where enterprise AI decisions actually get made.
Watch for: API pricing confirmation and context window specifications from OpenAI’s primary documentation. Watch also for how the “High” Preparedness Framework classification affects enterprise procurement conversations in regulated industries. OpenAI’s full announcement should be consulted directly when the page is accessible for complete technical specifications.