ChatGPT's free tier goes unlimited — OpenAI plays defense at scale
OpenAI removed free-tier text chat limits and promoted GPT-5.6 Luna to the default model
The three lines
- GPT-5.6 Luna becomes the default for Free and Go users, with unlimited text chats
- A new Think button brings deeper reasoning to free users — rolling out the week of August 10
- Tool limits (files, images) remain — this is a defense of the default position, not charity
Key questions
- What exactly do free users get
- Three things: the default model upgrades to GPT-5.6 Luna, text conversations lose their caps, and a Think button unlocks deeper reasoning on hard questions. File uploads, image generation and other tools keep their limits, and abuse guardrails still apply to the unlimited chats.
- When does it roll out
- The model swap lands the week of the announcement; unlimited chats and the Think button follow the week of August 10. Paid tiers get an updated GPT-5.6 Sol retuned for everyday conversation, plus a slider controlling how much thinking the model applies to each answer.
- Why would OpenAI give this away
- Because the default position is the asset. DeepSeek and LG's K-EXAONE are pushing free, downloadable open models; Google is embedding Gemini into Android. Being the AI people open without thinking is worth more than marginal subscription conversions. OpenAI claims Luna makes 62% fewer factual errors than the previous free default — a self-reported figure.
In the AI industry, "free" is not charity — it is trench-digging. OpenAI announced Thursday that ChatGPT's free tier is losing its text-chat limits, with the default model upgraded to GPT-5.6 Luna and a new Think button offering deeper reasoning to non-paying users. The timing, a week after DeepSeek and LG flooded the zone with open-weight releases, does not look accidental.
1. What changes — the ceiling comes off
Three changes for Free and Go users. The default model becomes GPT-5.6 Luna, which OpenAI says produces 62% fewer factual errors than the previous free default (its own benchmark — treat accordingly). Text conversations lose their message caps. And the Think button, previously a paid affordance, lets free users spend more reasoning on hard questions.
It is not everything-for-nothing. File uploads, image generation and other tools keep their usage limits, and the unlimited chats sit behind abuse guardrails. The subscription pitch shifts toward tools and the top models: the same announcement retunes GPT-5.6 Sol for Plus and Pro subscribers and adds a slider that sets how much thought the model invests per answer, from quick replies to extended-effort work.
2. The rollout, tier by tier
| Item | Free / Go | Plus / Pro |
|---|---|---|
| Default model | GPT-5.6 Luna (this week) | GPT-5.6 Sol update |
| Text chats | Unlimited (week of Aug 10) | Already unlimited |
| Deep reasoning | Think button (week of Aug 10) | New thinking-effort slider |
| Files, images, tools | Limits remain | Tier allowances |
| Guardrails | Anti-abuse limits apply | — |
Place the table on the industry map and the intent is legible. Free, downloadable open models — DeepSeek's V4-Flash, LG's K-EXAONE 2.0 — keep raising the floor of what costs nothing. Google is wiring Gemini into Android as a default. OpenAI's counter is not a benchmark chart; it is accessibility. The seat being defended is "the first AI hundreds of millions of people open without thinking" — an asset the company evidently values above the marginal subscription upgrade.
3. What remains
Two tests start next week: whether unlimited chats survive contact with real traffic, and whether the anti-abuse guardrails quietly become the new cap. Regional rollout timing, including Korea, is unannounced. The open-model offensive this move answers runs in "LG open-sources a 750B model" and "What 'open-source AI' actually means"; the safety-incident backdrop is in "AI labs disclose sandbox escapes."