Key Takeaways
- OpenAI improved effective usage for Codex and Work by reusing context across subagents, keeping context cached longer and improving request routing to raise cache hits.
- Codex is aimed at generating, reviewing and fixing code, while Work handles research, summaries and documents for business tasks.
- The update means you can complete more tasks before throttling, and it also includes improvements to GPT-6 Pro.
- There is no dedicated page for this update yet; check the official site, openai.com, for current limits and availability.
Quick Facts
| Type | AI coding and business task tools |
|---|---|
| Devices | Web-based; runs in the browser |
| Price | Check official site for current plans |
| Languages | Check official site for supported languages |
| Highlight | Higher effective usage via context reuse and caching |
| Learning curve | Low for basic prompts; moderate for advanced workflows |
| Category | Dev & Web Tools |
| Creator | Tibo |
| Official site | openai.com ↗ |
| Heat | 🔥 1.5M Heat |
| Last updated |
What OpenAI Codex & Work Actually Do
OpenAI Codex and Work are two AI tools from OpenAI designed for different but related jobs. Codex focuses on code: it can generate new code, review existing code and help fix bugs. Work is built for business tasks such as research, producing summaries and drafting documents. Both are accessed through OpenAI's platform, and both are aimed at people who want to hand off repetitive or time-consuming work to an AI assistant.
The recent update, announced by OpenAI's Tibo, is not about adding a new feature you can see in the interface. Instead, it improves the effective usage you get from the tools you already have. That means you can accomplish more before running into usage limits. The improvements come from three technical changes: reusing context across subagents, keeping context cached longer, and better request routing that increases cache hits.
This matters because usage limits are one of the main friction points for anyone using AI tools heavily. If you are a developer running Codex on a large codebase, or a team using Work to process documents all day, hitting a throttle can stop your workflow. By making the system more efficient behind the scenes, OpenAI is effectively giving you more headroom without changing your plan or the interface.
- Codex for code
Codex generates, reviews and fixes code. It is suited to developers who want to speed up implementation or catch issues in existing files.
- Work for business tasks
Work automates research, summaries and document creation. It is aimed at knowledge workers who need to process information quickly.
- Usage improvements
The update increases effective usage through context reuse, longer cache retention and better routing, so you can complete more tasks before throttling.
- GPT-6 Pro improvements
The announcement also mentions improvements to GPT-6 Pro, though specific details are not provided in the notes; check the official site for more.
How Context Reuse, Caching and Routing Boost Your Limits
To understand why this update matters, it helps to know a little about how large language models handle requests. When you send a prompt, the model processes the context you provide, which can include previous messages, files or instructions. Processing that context takes computing resources. If the same context is needed again, caching it means the system can reuse it instead of recalculating from scratch. That saves time and resources, which is why cache hits are valuable.
Subagents are smaller AI agents that a main agent can spawn to handle parts of a task in parallel. In the past, each subagent might have started with a fresh context, meaning the same background information had to be processed multiple times. The update changes this by reusing context across subagents. If you have a main agent working on a coding task and it spawns subagents to handle different files, those subagents can now share the same context instead of each rebuilding it.
Keeping context cached longer means that if you return to a task after a short break, the system may still have your context ready, so you do not pay the cost of reprocessing it. Improved request routing means that when you send a request, the system directs it to a server or model instance that already has relevant context cached, increasing the chance of a cache hit. Together, these changes reduce the overhead per task, which translates into more effective usage before you hit a limit.
- Context reuse across subagents
Subagents now share context instead of each starting fresh. This reduces redundant processing when tasks are split into parallel subtasks.
- Longer cache retention
Context stays cached longer, so returning to a task after a pause is more likely to benefit from cached data rather than reprocessing.
- Improved request routing
Requests are routed to increase cache hits, meaning the system tries to send your request where relevant context is already available.
- Net effect: more tasks per limit
Because each task consumes fewer resources, you can complete more work before throttling, effectively raising your usable capacity.
Getting Started with Codex and Work: A Practical Workflow
To start using Codex or Work, you need an OpenAI account and access to the platform. Log in to OpenAI, then choose the tool that fits your task. For Codex, you might paste in a code snippet and ask it to generate a function, review a file for bugs or suggest improvements. For Work, you might provide a topic or a set of documents and ask for a research summary or a draft. The interface is conversational, so you can iterate on results.
A practical workflow for Codex is to start with a clear, specific request. Instead of asking it to fix your whole project, give it one file or one function at a time. If the task is large, break it into steps and let Codex handle each step. You can also ask it to explain its changes so you can verify them. Because subagents can now reuse context, tasks that involve multiple related files may benefit more from the update, so structuring your work into related chunks can help.
For Work, the key is to provide enough context for the AI to understand your goal. If you need a summary of a long document, paste the document or upload it if the interface allows. If you need research, specify the questions you want answered and the format you want. You can then ask Work to refine its output. Since caching improvements mean repeated or related requests are more efficient, batching similar tasks together can help you get more done within your limits.
- Be specific with prompts
Clear, narrow requests produce better results and use fewer resources than vague, broad ones. For Codex, target one file or function at a time.
- Batch related tasks
Because context is cached longer and reused, grouping related requests can increase cache hits and reduce overhead.
- Verify Codex output
Always review generated or modified code before using it. Ask Codex to explain its reasoning if you need to understand the changes.
- Use Work for structured output
Tell Work the format you want, such as bullet points or a summary, to get usable results faster and reduce back-and-forth.
Who Benefits, How It Compares and What to Watch For
The people who benefit most from this update are heavy users of Codex and Work. If you are a developer who uses Codex throughout the day, or a team that relies on Work for research and documentation, the increased effective usage means fewer interruptions. Casual users may notice less of a difference, but anyone who has hit a throttle mid-task will appreciate the extra headroom. The update also mentions improvements to GPT-6 Pro, which may affect performance for users of that model.
Compared with other AI coding assistants and business task tools, OpenAI's offering is distinguished by its integration with the broader OpenAI platform and its use of subagents and caching. Many competing tools also use caching and context management, but the specific combination of reusing context across subagents and routing for cache hits is a notable engineering effort. The result is not a new feature you can toggle, but a behind-the-scenes efficiency gain that changes how much you can accomplish.
There are limitations to keep in mind. The notes do not specify exact new limits, so you should check the official site, openai.com, for current details on usage and plans. The update is described as improving effective usage, not removing limits entirely. Also, the availability note says there is no dedicated page for this update yet, so information may be spread across the official site or announcements. As always, verify details on the official site before relying on them for planning.
- Best for heavy users
Developers and teams who use Codex or Work daily will see the most benefit from the increased effective usage.
- Not a new interface feature
The changes are under the hood. You will not find a new button; you will simply be able to do more before hitting limits.
- Check official site for specifics
Exact usage limits, plan details and GPT-6 Pro improvements are not in the notes. Confirm on openai.com.
- Compare with alternatives
Other AI coding and business tools exist. Compare based on your workflow, but note that this update focuses on efficiency rather than new capabilities.
Step-by-Step Guide
Log in to OpenAI
Go to the official site, openai.com, and sign in to your OpenAI account. If you do not have one, you will need to create it. Access to Codex and Work may depend on your plan, so check what is available to you.
Choose Codex or Work
Decide whether your task is code-related or business-related. Use Codex for generating, reviewing or fixing code. Use Work for research, summaries and documents. You can switch between them as needed.
Craft a clear request
For Codex, provide the specific code or file and state what you want done. For Work, provide the source material or topic and specify the output format. Clear requests lead to better results and more efficient use.
Iterate and refine
Review the AI's output and ask for changes. If the task is complex, break it into smaller steps. Because context is reused and cached, related follow-up requests may be more efficient than starting over.
Monitor your usage
Keep an eye on how much you are using the tools. The update increases effective usage, but limits still exist. Check the official site for details on your plan's limits and how they are measured.
Frequently Asked Questions
What exactly changed with OpenAI Codex and Work usage limits?
OpenAI improved effective usage by reusing context across subagents, keeping context cached longer and improving request routing to increase cache hits. This means each task consumes fewer resources, so you can complete more tasks before hitting a throttle. The change was announced by OpenAI's Tibo and also mentions improvements to GPT-6 Pro.
How do I access Codex and Work?
You access them through OpenAI by logging in on the official site, openai.com. Availability may depend on your plan or region, so check the official site for current access details. There is no dedicated page for this specific update yet, so look for information on the main OpenAI site.
Do these improvements cost extra?
The notes do not mention any price changes. The update is described as an improvement to effective usage, not a new paid feature. For current pricing and plan details, check the official site, openai.com, as prices and plans can change.
What are subagents and why does context reuse matter?
Subagents are smaller AI agents that a main agent can spawn to handle parts of a task in parallel. Reusing context across them means they do not each have to reprocess the same background information. This reduces overhead, which is why it increases the effective usage you get from Codex and Work.
Will this remove usage limits entirely?
No. The update increases effective usage, meaning you can do more before throttling, but limits still exist. The notes do not specify the new limits. For exact details on your plan's limits and how they are measured, check the official site, openai.com.
