Improving GPT‑5.6 Sol in ChatGPT—and expanding access to GPT-5.6 Luna for free users

OpenAI is tweaking its lineup again, and this time the changes actually favor the crowd not paying a monthly fee. GPT-5.6 Sol, the model doing heavy lifting inside ChatGPT, gets a consistency and accuracy pass — the kind of unglamorous update that doesn't generate headlines but quietly fixes the annoying stuff, like answers that wobble between sessions or confident wrongness on straightforward questions.

The bigger move is Luna. GPT-5.6 Luna is now open to free-tier users, and not in some throttled, three-messages-a-day way. OpenAI is framing it as unlimited for everyday chats, which is a meaningful shift from the usual free-tier playbook of dangling a good model just long enough to make you want the paid one.

This matters because the free-versus-paid gap in AI chatbots has been the industry's quiet tension for two years now. Every lab wants enough people using the free version to build habit and data, but not so much good stuff for free that nobody upgrades. Giving free users unrestricted access to a named, numbered model — not a stripped-down cousin — signals OpenAI thinks its moat is elsewhere: enterprise tools, agents, memory, whatever comes next in the Sol line.

Sol's accuracy improvements, meanwhile, are aimed at paying users and developers who've been quietly grumbling about inconsistency in longer or more technical exchanges. OpenAI hasn't published benchmark deltas here, so it's hard to say how large the jump actually is. But the direction — tighten the flagship, loosen the free tier — reads like a company optimizing for user growth over near-term monetization, at least for this round.