OpenAI is taking away probably the most irritating restrict free ChatGPT customers face: the cap on textual content messages. The corporate introduced the change on August 6. Open ChatGPT as we speak, although, and you’ll not discover limitless chats. They arrive later, and OpenAI is exact in regards to the timing in its personal announcement.
What’s already dwell
To this point solely the paid facet has modified. Since August 6, Plus and Professional subscribers get an up to date model of GPT-5.6 Sol in chat, together with a slider that units how a lot thought the mannequin places into a solution. Depart it low for on a regular basis questions, push it up for planning, analysis or coding. Instantaneous replies and pondering now run on the identical mannequin for Plus and Professional, so the shift in tone between the 2 modes is gone.
This model of Sol is proscribed to common chat. ChatGPT Work and Codex are unchanged.
What remains to be to return
Free and Go customers get the change in levels. GPT-5.6 Luna turns into the default mannequin over the course of this week, changing GPT-5.5 Instantaneous. Limitless textual content chats and the brand new Assume button observe within the week of August 10, in response to OpenAI. The Assume button offers Luna extra time on more durable questions, a lot because the slider does for paying clients.
So in the event you learn as we speak that the restrict is gone, verify the date. Readers already know the mannequin names from the GPT-5.6 launch in July, when Sol, Terra and Luna arrived collectively.
The place the boundaries keep
Limitless means plain textual content. File uploads, pictures and the remaining instruments maintain their current caps. OpenAI additionally states that the limitless chats are topic to abuse guardrails, so anybody pushing the system laborious ought to nonetheless count on a brake.
Fewer errors, OpenAI says
OpenAI cites figures from an inner analysis. It examined finance, medical and authorized questions that activate laborious element. Responses containing at the least one factual error had been about 62 p.c much less frequent with GPT-5.6 Luna than with GPT-5.5 Instantaneous, and about 68 p.c much less frequent with GPT-5.6 Sol. No impartial verify of these numbers exists to date. On pricing, OpenAI had already cut the cost of its cheapest model.
New boundaries for youngsters
OpenAI has additionally expanded its safeguards for customers beneath 18. The mannequin is skilled to keep away from romantic roleplay and age-restricted challenges, and to not current itself as an alternative to actual relationships. Age-appropriate boundaries cowl sexual content material, consuming issues and physique picture, age-restricted items, harmful actions and graphic violence. The system card has the small print.
I write about IT safety and synthetic intelligence. Cybersecurity is one in all my common matters, amongst others for Golem.de. At Notebookcheck, my focus is on sensible AI for on a regular basis customers. I’ve labored hands-on with AI for years, most just lately with a give attention to AI brokers and their use in on a regular basis work, and I take a look at instruments and fashions myself slightly than simply studying about them.
