DevDay 新品实测：Dots 好评，6.1 Sol 不如 Astra

Gegam @Gegam245074原推发布于 2026.09.30

在 X 上打开原推

Yesterday I made a lot of negative posts about OpenAI. But today I tried out their new DevDay releases, and here's what I want to say: 1. Dots. Definitely better than Grok bot, Muse, Open Claw, and Hermes. We can complain about GPT models all we want, but they're still among the best, alongside Anthropic's. And the new product vision combined with the GPT Astra models delivers an experience on a much higher level than the rest of the competition. Unfortunately, I haven't come up with a use case for myself yet. But Dots is definitely a good product. I'll share a more detailed opinion once I've used it a bit longer 2. GPT-6.1 Sol. It's now much, much better than GPT-6 Sol. But the fact that they released it just a week later speaks to OpenAI's lack of foresight and poor understanding of the market. It still feels like a GPT model, with all its shortcomings. 6.1 Sol doesn't reach the level of Astra or Opus 5.5. It's more like something at the Sonnet 5.5 level, but with worse taste. And for the record, Sonnet 5.5 turned out really great. A more detailed review later 3. Decision API. I don't have access yet, but it'll definitely be better than Jev at a lot of things. Under the hood it uses a smarter Luna with vision. Neither Jev nor any other open classifier has that. As soon as I get access, I'll set it up for labeling the dataset for my project. Can't wait 4. Codex in the cloud. Still fairly limited in what it can do. On the one hand, it's a great thing and a logical evolution for coding agents; on the other, it's not suitable for everything. It's convenient for small, targeted tasks that depend on your repository. But you can't move your project's entire context and environment to the cloud and leave it running overnight 5. Agents API. The only use case I see is an extended version of Codex in the cloud, where you can configure your own remote environment for agents to work in. OpenAI themselves couldn't even properly demonstrate any decent usage scenarios. It's obvious they built this API alongside Codex in the cloud and decided to offer it as an extra service just in case 6. Code review. I've said many times that GPT models have no equal when it comes to code review, analysis, and bug hunting. The new Code review feature makes this process even better, especially for engineers who like reading code and following changes 7. Plugin extensions. This should be seen as an extension of the previous Codex plugin system, now with a visual component added. The new mini-app platform for Codex will be especially useful for small businesses that operated as API wrappers around ChatGPT and Claude. Now they don't have to bother with reselling tokens, and the product's price, user acquisition cost, and overall value to the consumer will all improve. Everything will now be included in a single subscription 8. ChatGPT Space. Surprisingly, I liked the implementation. I'll definitely be using it a lot. What I liked most is that they fully merged Space into a single service with your ChatGPT data. All the documents, files, and images you've used or created with your agent are organized in a convenient way, and you can share them with other people. It could definitely become a Notion replacement. Anthropic should definitely take Claude Design in this direction. It's not as convenient yet and isn't natively integrated into Claude Conclusions We've criticized OpenAI a lot for a lack of innovation, but setting that aside, they showed some good services. In terms of execution and product development, they're done better than the competition's and will definitely be in demand But there was no need to whip up such cosmic hype or call it an epoch-defining DevDay. These services and updates don't deserve that level of hype. An announcement on X would have been enough, like they do for new model releases, or a few short videos on YouTube