aiexpert
Home / News / Brief
Breaking · Aug 07, 2026, 07:05 AM · 7 sources

OpenAI moves GPT-5.6 Luna to free tier, removes text-chat limits for 1B+ users

OpenAI announced on August 6, 2026 that GPT-5.6 Luna becomes the default model for ChatGPT Free and Go users this week, with unlimited text chats rolling out the week of August 10. Free and Go users will also receive a new Think button that lets Luna allocate more reasoning time to harder questions. The change affects ChatGPT's 1 billion weekly users and represents a substantial upgrade from GPT-5.5 Instant; Luna reduces factual errors by 62% versus its predecessor on internal evaluations of financial, medical, and legal prompts. File uploads, image generation, and other tools remain limited on free accounts.

For paid subscribers, OpenAI retuned GPT-5.6 Sol specifically for ChatGPT conversations. The updated Sol unifies Instant and deep-reasoning responses into one model with a new effort slider (web, mobile, desktop) letting Plus and Pro users adjust reasoning depth per query—from fast answers to thorough analysis for coding, planning, writing, and research. OpenAI claims the retune delivers 68% fewer factual errors than GPT-5.5 Instant on the same benchmark suite. This retuning applies only to ChatGPT Chat; the versions of Sol powering ChatGPT Work and Codex remain unchanged.

For practitioners, the move is a bet that free-tier quality and distribution matter more than monetization friction. ChatGPT now competes on substance rather than access—the paid moat shrinks to the Sol reasoning slider and speed, and rivals (Gemini, Meta AI, Claude via Anthropic's apps) must now match Luna's factual accuracy on the free tier to stay competitive. OpenAI recently cut Luna API prices 80% and Terra 20%; this consumer rollout extends that cheap end of the GPT-5.6 family to the largest surface. Limits on abuse guardrails apply; the Think button and unlimited chat remain subject to safety guardrails and monitoring.

Sources

Everything this brief rests on
  1. 01 Primary source axios.com
  2. 02 axios.com axios.com “Unlimited text chats will be given to free users and those who subscribe to the low-cost Go plan.”
  3. 03 axios.com axios.com “OpenAI says that responses featuring at least one factual error were 62% less common with Luna opposed to the previous default model, GPT‑5.5 Instant.”
  4. 04 aiweekly.co aiweekly.co “OpenAI extended a meaningfully better model to all 1 billion ChatGPT free users—GPT-5.6 Luna posts 62% fewer factual errors than its predecessor”
  5. 05 cybersecuritynews.com cybersecuritynews.com “A new slider lets Plus and Pro users manually control how much computational effort ChatGPT dedicates to each query”
  6. 06 neowin.net neowin.net “this updated model would make fewer mistakes compared to the previous model when responses depend on dates, numbers, sources, rules, or assumptions.”
  7. 07 axios.com axios.com “68% fewer responses containing at least one factual error from the updated GPT‑5.6 Sol versus GPT‑5.5 Instant”