In the AI circle these past few days, it’s basically gods fighting it out.

Claude just reset its quota, and OpenAI immediately followed up with a GPT-5.6 Sol + Work mode reset.
I love this kind of corporate warfare— the harder it gets, the cheaper productivity becomes.
OpenAI’s current strategy is very straightforward:
In its blog, it keeps mentioning Claude, essentially telling everyone that when capabilities are roughly comparable, the cost of Sol is only one quarter of Fable 5, and Terra is even just one sixteenth.
Don’t be fooled by the fact that Sol hasn’t fully surpassed the previous generation or its rivals in some extreme math problems and pure coding benchmarks yet— in real delivery, it has made massive optimizations.
It can write its own scheduling scripts, decide when to use how much compute, and spin up multiple Agents in parallel.
That’s the handoff of control.
Before, we had to figure out how to write prompts to make the AI behave— now you just give it a goal, and it breaks down the steps and handles the rest on its own.
The product format has also changed dramatically:
1. The standalone Codex app is gone.
2. The Atlas browser has become a plugin.
3. The focus is now on ChatGPT Work.
One more detail: Sol’s pricing is actually redefining “cheap.”
It’s not just a lower unit price— it saves money by reducing redundant outputs.
When people provided feedback, after switching modes, the number of steps for users to build an application dropped by 25%, and the probability of tasks getting stuck fell by 15%.
This kind of end-to-end delivery capability will make a lot of AI wrapper companies that used to have no barrier feel uncomfortable.
In the future, the real barrier will only be practice.
How much real business experience you have in this field determines what kind of Agents you can “train and tune.”
The age of conversation is officially over; the era of delegation begins.