Updated
Updated · The New Stack · Aug 6
OpenAI Updates GPT-5.6 Sol in ChatGPT, Cuts Factual Errors by 68%
Updated
Updated · The New Stack · Aug 6

OpenAI Updates GPT-5.6 Sol in ChatGPT, Cuts Factual Errors by 68%

3 articles · Updated · The New Stack · Aug 6

Summary

  • Consumer ChatGPT now runs an updated GPT-5.6 Sol, while the Sol versions used in Codex and ChatGPT Work remain unchanged.
  • A new slider for Plus and Pro users on web, mobile and desktop lets them choose how much reasoning time ChatGPT uses, replacing separate quick-answer and deeper-thinking model choices.
  • OpenAI said the updated Sol produced answers with at least one error 68% less often than GPT-5.5 Instant on internal financial, medical and legal tests, but it did not publish prompts or enough detail to reproduce the results.
  • The company also compared the new ChatGPT Sol with GPT-5.5 Instant rather than the prior Sol version, leaving teams to test their own prompts to judge whether the upgrade improves real-world performance.

Insights

What hidden catches remain for free users now that OpenAI claims text conversations are entirely unlimited?
Will sliding ChatGPT's new reasoning control to the maximum secretly drain your paid usage quota in minutes?
Could forcing ChatGPT to think harder with the new slider actually cause more hallucinations on simple tasks?