engblogs

Case study: swapping in shared code-exploration tools for Copilot code review initially increased cost until tool instructions were rewritten for a diff-first reviewer workflow, delivering ~20% lower average review cost without sacrificing quality.

By aligning tool usage with a reviewer's workflow—start from the diff, narrow with grep/glob, read with view, and anchor in benchmarks—the team cut Copilot code review’s average cost by ~20% without sacrificing quality.

A concise overview of OpenAI's evolving Bio Bug Bounty program, outlining universal jailbreaks, biosafety for frontier AI models, and the shift from GPT-5.5 to GPT-5.6 with higher rewards and an ongoing scope.

GPT-5.6 frontier models scale intelligence and efficiency across coding, knowledge work, cybersecurity, and science with multi-agent coordination, cost-aware performance, and layered safeguards.

Explores ChatGPT Work, an autonomous agent that coordinates multi-step workflows across apps and documents using Codex and GPT-5.6, with plugins and Sites to transform inputs into deliverables and sustain long-running projects.

OpenAI expands its Bio Bounty into a private program focused on universal jailbreaks to strengthen biosafety in frontier AI models, with higher rewards and a defined 2026 scope.

GPT-5.6 becomes the preferred model in Microsoft 365 Copilot, enabling higher-quality drafting, analysis, and collaboration across Word, Excel, PowerPoint, and Cowork with more efficient tokens and on-demand AI capabilities.

GPT-5.6 delivers frontier-scale intelligence with scalable efficiency across Sol, Terra, and Luna, introducing ultra multi-agent workflows and enhanced safeguards to boost performance while lowering token-based costs.

Overview of OpenAI's GPT-5.5 Bio Bug Bounty program, evolving into GPT-5.6 and focusing on universal jailbreaks to probe biosafety safeguards in frontier AI biology, with increased rewards and a rolling application process.

ChatGPT Work extends ChatGPT into an enterprise-ready agent that orchestrates multi-step workflows across apps and files, powered by Codex, GPT-5.6, plugins, and Sites to turn ambitious goals into finished work.

GPT-5.6 becomes the preferred model in Microsoft 365 Copilot, delivering higher-quality, more efficient AI-assisted productivity across Word, Excel, PowerPoint, and collaborative work.

A technical overview of GPT-5.6’s frontier-model lineup (Sol, Terra, Luna), highlighting unprecedented cost efficiency, multi-agent coordination, and robust safeguards, plus tooling innovations like Programmatic Tool Calling that accelerate coding, research, and cybersecurity tasks.