OpenAI has announced an important update for GPT-5.6 Sol, addressing one of the biggest concerns raised by ChatGPT Work and Codex users since the model’s launch. After investigating reports that Sol was consuming usage limits much faster than expected, the company has introduced several backend optimizations that should make users’ quotas last around 18% longer during typical workloads.
At the same time, OpenAI has reset usage limits for all ChatGPT Work and Codex users, giving everyone a fresh start while rolling out the improvements.
The update also marks the return of the five-hour usage limit, which had been temporarily paused while engineers investigated the unexpected increase in token consumption.
GPT-5.6 Sol Wasn’t Using Fewer Limits—It Was Working Harder
According to OpenAI, subscription plans were never reduced. Instead, GPT-5.6 Sol naturally consumed more resources because of how it approaches complex tasks.
Unlike previous models, Sol is designed to:
- Work longer on difficult problems
- Make additional tool calls automatically
- Coordinate multiple tools and subagents
- Handle more sophisticated reasoning workflows
These improvements significantly increase the model’s capability, but they also resulted in higher-than-expected token usage in certain scenarios.
Why GPT-5.6 Sol Used More Tokens
OpenAI shared several reasons behind the increased usage.
More complex reasoning
GPT-5.6 Sol continues working longer before producing an answer, especially for difficult coding and research tasks.
Smarter tool usage
The model performs more web searches and executes more tool calls than previous generations to improve answer quality.
Parallel code mode
Programmatic tool calling, also known as Code Mode, allows Sol to:
- Run multiple tool calls simultaneously
- Continue working while waiting for tool responses
- Coordinate several workflows in parallel
While this dramatically improves productivity, it also generated:
- More responses per conversation
- Higher cached input tokens
- Greater token consumption overall
OpenAI Has Optimized Sol’s Efficiency
After analyzing real-world usage, OpenAI says it has already deployed multiple improvements.
The company expects:
- Around 18% longer usage for typical users
- Some users to see even larger efficiency gains immediately
- Better handling of web searches
- Improved efficiency while waiting for tool responses
- Continued optimization of Code Mode
These backend changes reduce unnecessary token consumption without sacrificing GPT-5.6 Sol’s advanced capabilities.
Power Users Were Hit Hardest
Interestingly, OpenAI says the problem was not evenly distributed.
Most users actually experienced reasonable token efficiency.
However, developers and power users running:
- Complex coding projects
- Multi-agent workflows
- Heavy research sessions
- Extensive web searches
were much more likely to exhaust their usage limits quickly.
OpenAI admitted it focused heavily on average and median usage before launch, which meant some high-end usage patterns were overlooked.
OpenAI Acknowledges the Issue
The company also acknowledged that it should have recognized the problem sooner and communicated more transparently.
OpenAI explained that GPT-5.6 Sol represents a major leap forward in capability, but improvements in intelligence do not always arrive alongside equal gains in efficiency. Some optimization challenges only became clear after millions of real-world interactions.
The company says it will continue refining Sol’s efficiency while keeping users informed as additional improvements roll out.
What This Means for ChatGPT Work and Codex Users
For developers and professionals who rely on GPT-5.6 Sol every day, this update should noticeably improve the experience.
The biggest changes include:
- Â Usage limits reset for ChatGPT Work and Codex users
- Â Approximately 18% better usage efficiency
- Â Five-hour usage limit restored
- Â Reduced unnecessary token consumption
- Â More efficient Code Mode and tool calling
- Â Continued backend optimizations planned
These improvements should allow users to complete longer coding sessions, research projects, and multi-tool workflows before reaching their usage limits.
Final Thoughts
GPT-5.6 Sol remains one of OpenAI’s most capable AI models, particularly for coding, research, and complex reasoning tasks. While its advanced workflows initially resulted in higher token usage for some users, OpenAI’s latest optimizations aim to strike a better balance between capability and efficiency.
With usage limits reset, an estimated 18% improvement in efficiency, and the five-hour limit returning, ChatGPT Work and Codex users should now experience longer sessions and better overall value as OpenAI continues refining the model.
