Oleksandr Kuzmenko, EditorPublished 6 min readAI-assisted · editor-reviewed
TL;DR
- ChatGPT Chat and Work tabs are consolidating into a single project interface in the desktop client.
- Codex and ChatGPT will now share one token allowance pool, raising rate-limit friction for intensive coders.
- Anthropic implemented a matching unification between Claude chat and Cowork earlier in September.
Key facts
- Rollout Phase
- Select users initially for feedback before wider release
- Codex UI Status
- Remains separate experience in ChatGPT desktop app
- Token Allocation
- Single shared allowance pool across ChatGPT and Codex
- Precedent
- Anthropic merged Claude chat and Cowork in September
Unified Canvas for Files, Web Actions, and Plugins
The ChatGPT desktop application is streamlining its dual-tab architecture. Previously divided into distinct Chat and Work tabs alongside a dedicated Codex environment, the new layout merges discussion and execution into a single canvas. Users can feed repositories, documents, and web plugins directly into the prompt stream, instructing the model to scaffold websites, generate slide decks, or trigger browser-based actions without context fragmentation.
Shared Token Allowance with Codex
The structural trade-off lies in quota pooling. OpenAI confirmed that ChatGPT desktop usage and Codex will now draw against identical token allowance limits. While casual users may not notice the difference, intensive vibe coders and tech leads running frequent Codex refactoring loops will hit usage boundaries earlier:
- Merged Quotas: Desktop chat prompts and Codex executions count against the same period limit.
- Reset Delays: High-volume sessions with frontier models will trigger rate-limiting cooldowns more frequently.
- Codex Separation: The Codex interface remains a distinct panel within the desktop application, but without dedicated capacity.
Strategic Alignment with Claude Cowork
This consolidation mirrors Anthropic's move in September, which merged Claude chat and Cowork tabs into a unified view. For developers, unified workspaces reduce copy-paste friction, but shared pools mean agent tasks directly cannibalize interactive brainstorming capacity.
When to use
- Building end-to-end web prototypes where prompts directly create and edit workspace files
- Streamlining workflows that require switching between browsing plugins and file review
When NOT to use
- Heavy automated batch refactoring in Codex that risks exhausting your interactive chat quota
- Teams relying on strictly isolated rate limits between IDE coding and general conversational research
What to do today
- Separate mission-critical Codex development from exploratory ChatGPT prompts to prevent unexpected lockout.
- Evaluate API keys or secondary seats if your daily throughput exceeds the unified desktop quota.
Step names are taken from What to do today.
Step names are taken from What to do today. Separate mission-critical Codex development from exploratory ChatGPT prompts to prevent unexpected lockout.. Evaluate API keys or secondary seats if your daily throughput exceeds the unified desktop quota.
- Separate mission-critical Codex development from exploratory ChatGPT prompts to prevent unexpected lockout.
- Evaluate API keys or secondary seats if your daily throughput exceeds the unified desktop quota.