Two New AI Models Are Coming — But Only One Works Today

Two of the biggest AI companies are in a quiet race again. Anthropic has an unreleased model called Claude Opus 5.2 that people say is being tested behind the scenes inside Claude Code — Anthropic's tool that writes computer code for you. OpenAI's GPT-6 family is different: it's already shipped, and you can use it today. So this isn't really a fair fight yet, because only one of them officially exists.

What do testers actually say about the hidden Claude model? They describe it as faster than the previous version, producing cleaner work on the first try, and — the big one — able to keep working on long coding jobs without someone constantly typing "continue." One developer described it as a "gauntlet loop": the AI writes code, tests it, fixes it, and repeats until the job is done. A tester claimed it's "on par with" GPT-6 Astra, OpenAI's higher-end model, and better at frontend design and 3D graphics. That's the closest thing to a real comparison anyone has, and it's one person's impression, not a score.

On the OpenAI side, there's at least some documented detail. GPT-6 Sol, the everyday member of that family, is reported to handle 15-minute 3D tasks using around 60,000 tokens (tokens are roughly word-sized chunks of text the AI reads and writes). In plain terms, it can work on a video-game-style 3D scene for a full quarter of an hour in one stretch without losing the thread.

Here's the catch, and it's a big one. Claude Opus 5.2 has no official model card, no published price, no known context window, and no benchmark numbers at all. There's even chatter that Anthropic skipped version 5.1 entirely — but that's inferred from a leaked name, not announced. If you're choosing an AI tool for your business today, GPT-6 is the only option you can actually buy, budget for, and rely on. The rest is a promising signal, not a product.