Observed facts: OpenAI’s October 7 system card says GPT-6 Sol and GPT-6 Luna replace GPT-5.6 Sol and GPT-5.6 Luna in ChatGPT, while Codex and ChatGPT Work remain on earlier versions. It reports that GPT-6 Instant begins answering 44% sooner than GPT-5.6 Instant on web-search questions. OpenAI also reports improved resistance to multi-turn jailbreaks and classifies the October models as High capability in cybersecurity and biological and chemical domains.
Mechanism: A first-token or answer-start improvement can reduce perceived waiting, while deployment-specific versions mean the headline model name does not guarantee every OpenAI surface is running the same release. The system card makes those boundaries visible alongside the company’s safety evaluations.
Why now: The update is framed as a broad ChatGPT rollout, but the stated exceptions for Codex and Work make product-surface tracking important now.
Implications: Teams should verify which model version serves each workflow before attributing latency or safety changes to “GPT-6” generally. The reported speed gain applies to web-search questions and answer start, not all tasks or completion time.
Evidence boundary: These performance and safety statements come from OpenAI’s own deployment-safety publication; the supplied evidence does not establish independent replication or generalize the 44% figure beyond the specified comparison.
Watch condition: Watch for confirmation that Codex and ChatGPT Work move to the October versions, and for comparable measurements on other task types.