OpenAI announced major upgrades to ChatGPT on August 6, 2026, removing daily text chat limits for Free and Go tier users starting next week. The company is deploying GPT-5.6 Luna as the new default model for free tiers this week, alongside a new Think button for complex queries. Meanwhile, Plus and Pro subscribers receive an upgraded GPT-5.6 Sol model and a thinking slider for granular reasoning control. OpenAI claims internal tests show factual errors are 62% less common in Luna and 68% less common in Sol compared to GPT-5.5 Instant.
Free tier text limits removed
- ▪The removal of ChatGPT text chat limits is scheduled to roll out to Free and Go tier users next week
- ▪OpenAI announced on August 6, 2026, that it is removing daily text-based chat rate limits for ChatGPT users on its Free and Go tiers
- ▪Separate usage limits will continue to apply to ChatGPT features including file uploads, images, voice chats, and image generation for Free and Go tiers
GPT-5.6 Luna default model
- ▪GPT-5.6 Luna is a lightweight model featuring a context window of 1,050,000 tokens and a knowledge base cutoff of February 16, 2026
- ▪OpenAI is making GPT-5.6 Luna the new default model for Free and Go ChatGPT users, replacing the previous GPT-5.5 Instant model
- ▪The rollout of the GPT-5.6 Luna model as the default experience for Free and Go ChatGPT users is scheduled to occur this week
GPT-5.6 Sol paid upgrade
- ▪The GPT-5.6 Sol model variants used for Codex and ChatGPT Work remain unchanged by the August 6, 2026 consumer update
- ▪OpenAI is upgrading the GPT-5.6 Sol model for ChatGPT Plus and Pro subscribers to deliver more focused, direct answers with tighter formatting
- ▪The upgraded GPT-5.6 Sol model and a new thinking slider are available to Plus and Pro subscribers starting August 6, 2026
Think button reasoning feature
- ▪ChatGPT Plus and Pro subscribers are receiving a new thinking slider to adjust the amount of reasoning the model applies to each response
- ▪OpenAI is introducing a new Think button for Free and Go ChatGPT users to allow the AI more processing time for complex questions
- ▪The new thinking slider for Plus and Pro users unifies the previously separate Instant and Thinking experiences into a single adjustable interface
Factual error reduction rates
- ▪OpenAI internal tests reported that factual errors in the upgraded GPT-5.6 Sol model were 68% less common than in GPT-5.5 Instant
- ▪OpenAI internal evaluations reported that factual errors were 62% less common with GPT-5.6 Luna compared to GPT-5.5 Instant
- ▪OpenAI has not published the evaluation dataset or had outside laboratories reproduce its reported 62% and 68% factual error reduction rates
Story comments
Loading comments…