OpenAI squeezes more usage out of Codex and Work
OpenAI just rolled out efficiency upgrades to Codex and Work, effectively giving users more mileage out of their usage limits. The improvements come from reusing context across subagents, caching data longer, and smarter routing to boost cache hit rates. Basically, your complex agentic workflows just got longer legs without extra cost.
If you use AI to write code or run multi-step tasks, you know how fast context windows and rate limits burn through.
On day 12 of its product update series, OpenAI rolled out a set of backend efficiency improvements to Codex and Work designed to boost the effective usage users get out of the tools.
Squeezing more out of subagents
The gains come down to three specific engineering fixes behind the scenes.
OpenAI is now reusing context across subagents instead of re-passing identical data at every step. It also keeps context cached for longer periods and revamped its request routing to dramatically raise cache hit rates.
Product summaries also noted accompanying updates to GPT-6 Pro for Pro subscribers, with users prompted to update their apps to feel the difference.
Why it matters
For power users chaining complex coding workflows or relying on subagents, this effectively stretches your quota further without paying extra.
We think this is a smart, unglamorous win. Multi-agent workflows are notoriously wasteful, constantly re-ingesting prompts and repo states across separate steps.
Instead of just throwing more raw compute or bigger context limits at the problem, optimizing cache hits and routing is where the real cost and speed improvements will come from as autonomous agents mature.
Sources (1)
- x.com(opens in a new tab)Originalx.com