techcrunch.com web signal

OpenAI Makes Luna Default for Free ChatGPT, Uncaps Text Chats

openai ai assistants ai-business

TL;DR

  • GPT-5.6 Luna becomes the default model for ChatGPT Free and Go users this week, replacing GPT-5.5.
  • Free and Go users get unlimited text chats plus a new Think button next week, though file, image, voice, and image-generation limits remain.
  • OpenAI's internal evals report factual errors 62% lower on Luna and 68% lower on Sol versus GPT-5.5-Instant.

OpenAI's update this week is the kind of move you make when the free tier is the front line of the assistant war rather than a demo. According to TechCrunch, GPT-5.6 Luna becomes the default model for ChatGPT's Free and Go users this week, replacing GPT-5.5, and next week those same users get unlimited text chats plus a new Think button that lets Luna spend more time on complex questions.

Paid subscribers get their own upgrade in parallel. Plus and Pro users are moved onto a retuned GPT-5.6 Sol aimed at everyday work like quick questions, web research, planning, writing, advice, and decisions, along with a thinking slider that adjusts how much reasoning the model puts into a given answer. OpenAI also says internal evaluations found factual errors 62 percent less common on Luna and 68 percent less common on Sol than on GPT-5.5-Instant, tested on financial, medical, and legal prompts that demand specifics.

The reason this matters is less about the model names and more about the pricing signal. Uncapping text chats on a tier that most rivals still meter is a bet that ChatGPT wins by being the default consumer surface rather than by squeezing free users toward a subscription. The trade OpenAI is making, quietly, is inference cost against retention and daily active users, and the way to read the Sol slider and the Think button is as the paid-tier justification once the free-tier message limit is gone.

The honest caveats are the ones any release like this earns. The 62 percent and 68 percent error-reduction figures come from OpenAI's own internal eval, not a third-party benchmark, so take them as a directional claim until independent testing lands. What the reporting does not spell out is where the new caps will fall on files, images, voice, and image generation, all of which remain metered, or how the unlimited text quota holds up if usage spikes.

If the accuracy numbers on Luna hold up in practice, the immediate beneficiaries are the free users and cost-sensitive markets who now get a materially more accurate default without paying, and the pressure moves onto Gemini and Meta AI to match the offer.