ChatGPT does not have one universal usage cap. OpenAI's current help article separates everyday chats, Thinking and Pro-model allowances, and usage in ChatGPT Work or Codex. Availability and reset behavior vary by plan and workspace, so old fixed-number summaries can become misleading quickly.

Current guidance: check the model and plan-specific allowance displayed in ChatGPT. OpenAI says Free and Go users have unlimited everyday text chats subject to safeguards, while separate limits apply to tools and some models. Managed workspaces and Codex have their own usage details.

What OpenAI documents now

The current OpenAI help article describes limits by product surface and model rather than one blanket message count. Treat the account UI and the linked article as the source for the allowance that applies to you; availability and policy can change.

Usage area Current documented behavior Why coders should care
Free and Go chats Unlimited everyday text chats, subject to safeguards; separate limits still apply to tools such as file uploads, image generation, voice, and data analysis. A chat allowance does not imply unlimited access to every tool or model.
Thinking models Availability and usage depend on plan; when a limit is reached, ChatGPT may offer a fallback model or show a reset time. Check the in-product notice before assuming a particular reset schedule.
GPT-6 Pro / GPT-5.6 Sol Pro OpenAI lists model-specific allowances that vary by plan; ChatGPT Work and Codex allowances are separate from Chat usage. Switching products or models does not necessarily share or reset an allowance.
Managed workspaces Business and Enterprise limits can depend on workspace settings and plan-specific rules. Confirm the workspace's policy and model access with its administrator.

A usage cap and project recovery are different problems.

A model allowance controls whether you can continue using that model or feature. It does not preserve the code changes, test results, failed attempts, or next step from an interrupted coding task. Keep those facts in the repository and a concise handoff, rather than relying on the conversation alone.

That is exactly where a local recovery layer matters. ShardStitch does not need the active ChatGPT thread to summarize itself; it reads the repo state, changed files, notes, and project memory from disk, then builds a smaller continuation package for the next session or tool.

The other big change

OpenAI's model lineup and plan limits change over time. Check the linked current usage article whenever you need exact model availability or reset details; this page avoids repeating fixed caps that may go stale.

For coders: the pain is not only the cap. It is the moment the cap interrupts context, interrupts momentum, and makes you rebuild the project state from memory.

What to do when you hit the ceiling

  1. Do not paste the whole transcript back into a new chat.
  2. Save your files and preserve the git state.
  3. Use ShardStitch to read the local working state and create a clean handoff.
  4. Restart in the same tool or move to another one with the context already shaped.

Why ShardStitch fits this moment

When usage limits change, the real loss is not the message count. It is the work you were carrying in the session. ShardStitch is built to recover that work from disk so you can keep going without reconstructing the project from scratch.

  • Reads git diff, changed files, and project memory from your machine.
  • Formats the handoff for the next tool.
  • Supports same-tool restart and tool-to-tool switching.
  • Stays local unless you choose to move the handoff yourself.
Privacy: no cloud sync, no telemetry, and no mandatory online account for recovery. Activation uses your purchase license key. Your code stays on your machine unless you decide to move it.

Sources