On July 9, OpenAI shipped three things at once: the new flagship model family GPT-5.6, a ChatGPT Work agent built for complex tasks, and the formal merger of its coding agent Codex into the ChatGPT desktop app. If you've been bouncing between ChatGPT, Codex CLI, and the API, this update essentially pulls the whole productivity chain into a single entry point.

What GPT-5.6 is: three tiers to choose from

GPT-5.6 isn't one model but a family, split into three tiers along the capability-cost curve, covering everything from heavy reasoning to high-frequency calls:

In the release notes, OpenAI stresses that Sol reaches the current state of the art on coding, knowledge work, cybersecurity, and research tasks, and outperforms Anthropic Fable 5; Terra also beats Fable 5, while Luna beats Claude Opus 4.8 — with all three costing roughly two-thirds of their comparison points.

Sol also introduces a top-tier setting called ultra, which coordinates multiple agents working in parallel on different subtasks — a fit for those complex projects that "take hours to run."

ChatGPT Work: an agent that can "stay at it for hours"

ChatGPT Work is the other highlight of this update. It is essentially an agent inside ChatGPT, purpose-built for tasks that span multiple apps and multiple steps to complete.

What it can do:

• Gather information from your apps and workflows
• Produce finished deliverables — spreadsheets, slides, documents, web apps
• Break complex projects into small steps and complete them independently

It is built on top of Codex — the agent OpenAI previously positioned for coding — now extended to productivity scenarios. The underlying model is GPT-5.6 Sol, and it supports generating materials based on templates and reference files.

Availability

• Desktop app: every plan (including the free tier) can use ChatGPT Work in the ChatGPT desktop app
• Web: now live for Pro, Enterprise, and Edu plans; Plus and Business plans open up over the coming days
• Codex merged into the desktop app: the coding agent is no longer a separate entry point — it's invoked directly inside the ChatGPT desktop app

💡 Tip: If you mainly used Codex CLI before, you can now switch to Codex workflows right inside the desktop app — no separate terminal needed.

How to pick a tier

• Writing complex code, security analysis, long-chain research → pick Sol (switch on ultra multi-agent parallelism when needed)
• Everyday coding, document processing, automation tasks → Terra is the most balanced on value
• Bulk calls, lightweight Q&A, token-cost-sensitive workloads → Luna; the $1/$6 price point suits scaled deployment

API and integration

All three GPT-5.6 tiers are live on the OpenAI API. Combined with Codex landing in the desktop app, the typical usage now looks like:

• Heavy development: run long tasks with ChatGPT Work / Codex on desktop
• Programmatic calls: route between Sol / Terra / Luna in the API by task complexity
• Team collaboration: members on Pro/Enterprise plans can share ChatGPT Work deliverables directly

Who it's for

• Developers and teams: for long-chain coding, cross-file refactoring, and multi-step research tasks, Sol + ultra saves a huge amount of manual coordination
• Knowledge workers: ChatGPT Work turns out finished spreadsheets/slides/documents, replacing part of the manual Office grind
• API callers: the three-tier pricing leaves room for cost optimization; high-frequency light tasks can sink down to Luna

References

• GPT-5.6 launch page (OpenAI)
• GPT-5.6 in ChatGPT (OpenAI Help Center)
• ChatGPT Work (OpenAI)
• OpenAI Announces ChatGPT Work and GPT-5.6 (Thurrott.com)