What Changed at OpenAI in 2026

If your ChatGPT knowledge is from spring 2026, a lot has shifted. Between late June and early November, OpenAI released two generations of models, rebuilt Voice mode, launched a hosted Agents API, shut down the Assistants API, introduced ChatGPT Work, and began replacing custom GPTs with plugins. This course walks through these changes in order: summer first, with the GPT-5.6 family; then fall, with the GPT-6 family; and finally, the deprecation schedule. It is current as of November 5, 2026.

Timeline

What Hasn’t Changed

The core principle still holds: at OpenAI, you choose a model tier, then a reasoning effort, request by request. This approach dates back to GPT-5, which replaced the o-series in 2025. GPT-5.6 extended it by adding the max effort, pro mode, and three named tiers: Sol, Terra, and Luna. OpenAI clarifies that the number denotes the generation, while Sol, Terra, and Luna are enduring capability tiers that can evolve independently. The GPT-6 family keeps this logic, with one key difference: its two most capable models, Astra and GPT-6.1 Sol, do not accept the none effort.

What Has Changed Since Summer

The models from July are no longer the ones OpenAI recommends for getting started. GPT-6 Astra is the most capable model, GPT-6.1 Sol comes close at a much lower cost, GPT-6.1 Terra targets speed, and GPT-6 Luna targets volume. For developers, the Agents API hands agent orchestration over to OpenAI, GPT-Live 1 enables full-duplex voice in the API, and the Decisions API delivers typed responses at scale. For users, ChatGPT Chat mode has run on GPT-6 since October 7, ChatGPT Work shares its usage with Codex, and custom GPTs are giving way to plugins.

This pace has a downside: several products are shutting down. The Assistants API has been gone since August 26. Agent Builder, the Evals platform, and reusable prompts stop on November 30, older image models on December 1, and gpt-5, o3, and custom GPTs on December 11. If your code or workflows depend on these, the final lesson provides the full schedule and an action plan.

How to Use This Course

Lessons 2 through 8 cover summer and the GPT-5.6 family. These models remain available via the API and even serve as the official replacement for the models retired on December 11; each lesson opens with a “Status as of November 5, 2026” box highlighting what has changed since. Lessons 9 through 11 present fall: the GPT-6 family, the new APIs, and ChatGPT updates. Lesson 12 consolidates the shutdowns and ends with the final quiz. Other OpenAI courses on this site still use GPT-5.6 identifiers in their examples: they still work, and you will know what to replace them with for a new project.

Your Turn

Inventory your dependencies. List the OpenAI models and products you use: in your code, your tools, and your internal workflows. For each, find its line in the timeline and classify it into one of three categories: still current, replaced by something newer, or closing soon. Keep this list handy: Lesson 12 will help you turn it into a migration plan.

Official sources: API Changelog · Deprecations (OpenAI API documentation) · ChatGPT release notes (OpenAI Help Center) · What’s new (ChatGPT documentation) · GPT-5.6 announcement (OpenAI, July 9, 2026)

Key Takeaways

• Two generations in four months: GPT-5.6 in July, then GPT-6 from September to October
• You still choose a model tier, then a reasoning effort; Astra and GPT-6.1 Sol do not accept none
• New APIs: Agents API, GPT-Live 1, and Decisions API; on the ChatGPT side: GPT-6, ChatGPT Work, and plugins
• Shutdowns to note: Agent Builder, Evals, and reusable prompts on November 30; custom GPTs, gpt-5, and o3 on December 11