OpenAI Just Launched GPT-5.6. It's Racing to Stop the Bleeding.

OpenAI shipped GPT-5.6 to the public today, and the timing says as much as the model does. Three weeks ago, ChatGPT's global web traffic share slipped below 50% for the first time since launch. Today's release is OpenAI's answer to that slide, and it arrives in a form the company has never tried before: not one model, but three.

Sol, Terra, and Luna are the new names, and they are not marketing flourish. Under OpenAI's own announcement, the "5.6" identifies the generation, while Sol, Terra, and Luna identify durable capability tiers that can each advance on their own cadence going forward. That's a real change to how OpenAI ships models, not just a naming refresh. Sol is the frontier reasoning tier, priced at $5 per million input tokens and $30 per million output, built for the hardest multi-step work with a new max-reasoning mode and an "ultra subagent" mode for delegating parts of a task. Terra sits in the middle at $2.50/$15. Luna is the cheap, fast option at $1/$6, positioned as the model most ChatGPT users will actually touch day to day.

The rollout tiers matter here. ChatGPT Pro, Team, and Enterprise subscribers get Sol and its advanced modes as the access unlocks. Plus subscribers, the largest single tier by user count, are getting Terra- and Luna-class models for daily use rather than the flagship reasoning tier. OpenAI is reserving its best model for the accounts that pay the most, which is not a new idea, but it's a more explicit version of it than the company has run before. The model also picked up more predictable prompt caching, with explicit cache breakpoints and a 30-minute minimum cache life, a detail aimed squarely at developers running GPT-5.6 through the API rather than casual chat users.

None of this happened in a vacuum. GPT-5.6 spent almost two weeks in a genuinely unusual limited preview: roughly 20 government-approved partner organizations got early API and Codex access starting June 26, gated behind a formal U.S. government safety review before the public got anywhere near it. That's a heavier-handed rollout process than OpenAI used for any prior GPT release, and it reflects a regulatory environment around frontier models that has tightened considerably since even six months ago. The company hasn't published details of what the government review actually checked, only that it happened before any partner outside that initial group of 20 organizations got access, which is itself a break from OpenAI's usual pattern of shipping to consumers first and answering regulatory questions after the fact.

Here's the part that actually explains why this launch matters beyond the spec sheet. ChatGPT still leads the AI chatbot market, at 53.9% of worldwide web visits across the seven largest generative AI chatbots. That's comfortably ahead of Google Gemini's 27.9% and Anthropic's Claude Sonnet 5 at 9.2%. But six months ago, in December 2025, ChatGPT held roughly 65% of that same market. A product can lose nearly 24 percentage points of share in a year and still be the outright leader; both things are true here, and neither one cancels the other out.

The two challengers aren't gaining the same way. Gemini's climb, from 527.7 million monthly web visits in May 2025 to 2.9 billion in May 2026, is real, but a meaningful chunk of it runs through Google Search, Android, and Workspace rather than people typing gemini.google.com into a browser on purpose. That's a different kind of growth than Claude's: Anthropic's model grew 855% year over year and 228% in a single quarter, the fastest expansion of the three, off a much smaller base and without a search engine or a phone operating system funneling traffic its way.

By user count, the picture looks even more lopsided than the traffic-share numbers suggest. ChatGPT still counts over 1.1 billion monthly users worldwide, against 662 million for Gemini and 245 million for Claude. Web traffic share measures something narrower than that, closer to which product people reach for when they have an open browser tab and a specific task, and it's the metric where OpenAI's lead is actually eroding fastest.

That's the real read on GPT-5.6. It isn't a response to Claude or Gemini catching ChatGPT in absolute size. It's a response to ChatGPT losing the tab-open, task-specific moments to a rival that increasingly has its own answer built into search and its own phone, and to a smaller rival growing off enthusiast word of mouth alone. A three-tier model lineup, with the strongest reasoning tier gated to the accounts paying the most and a cheap fast tier pushed to everyone else, reads like a company trying to hold two different fights at once: keep power users from drifting to Claude on hard tasks, and keep casual users from drifting to Gemini on convenience. Whether Sol, Terra, and Luna actually do either of those things won't be visible in this week's numbers. It'll show up, if it shows up at all, the next time someone runs this same web-traffic-share count in another six months.