Practitioners get a free Apache 2.0 model for local agentic workflows — tool calls, vision, multi-step reasoning, and failure recovery — with under-20GB memory footprint and up to 3× speedup via speculative decoding.
Why it matters: Open-weight agentic AI that runs on local hardware is now a credible deployment option where per-token API cost or data residency is a hard constraint.
The new tier removes the standard Business seat's five-hour usage limit; orgs signing up before August 20 get $100 in workspace credits per Premium seat added (up to five).
Why it matters: Teams that have been hitting standard ChatGPT Business usage limits now have a dedicated, budgetable tier to evaluate.
GPT-5.6-Cyber completes 95% of exploit-chain tasks in internal evals vs. 1.5% for standard GPT-5.6; Daybreak Blue opens frontier models to defenders, Daybreak Red gates the new model for vetted security teams.
Why it matters: OpenAI's dual-tier controlled-access model for high-stakes AI is a governance pattern likely to spread across regulated verticals as capabilities compound.