用户长文致谢 OpenAI：从 Opus 4.5 换到 Astra

Ishu Agrawal @ishuagra02原推发布于 2026.09.29

在 X 上打开原推

Dear Tibo @thsottiaux , Tomorrow's the day I've been waiting for since the last few months. The day when we see what y'all have been cooking. An event dedicated to paving the future of ChatGPT. A pitch on how it will transform people's lives for the better. Opus 4.5 was the first AI model I used for my work. It was fast, it was fun to talk to, and it got things done. But in its next few releases, the performance worsened, and the limits got tighter. I wanted to try something new. OpenAI offered me a free month trial to ChatGPT Plus, where I could access GPT 5.4 via Codex. I tried it out, and was amazed at how thorough the model was, along with its very generous usage. Then, GPT 5.5 came out and it took a huge leap forward in capability. On top of that, OpenAI increased my limits by 10x for ONE ENTIRE MONTH. I stopped using Claude. ChatGPT was everything I needed. My limits felt endless, more so with the resets, and it made me feel that OpenAI truly was delivering AI for everyone while Anthropic began catering to the elite. Soon after, GPT 5.6 came out, and Sol quickly became my most used model. Though it was very capable, it also began draining my limits more quickly. I posted about this, and many OpenAI engineers started looking into the usage issues. Eventually, you confirmed various bugs with the harness that the team had resolved, and it once again confirmed to me how vocal the team is with the community. It made me feel that my voice was heard, and that OpenAI takes customer feedback seriously. Earlier this month, the team released Astra, and it showed me a level of intelligence I had never seen before in a model. It led to many fun experiments, and I was amazed the team made this model available even in the $20 plan. Unfortunately, Astra was so expensive that it was barely usable in the Plus plan, and even in the Pro plan, I could drain my limits within 2 days if I tried. It felt weirdly unlike OpenAI, and I was forced to jump to Opus 5.5 when it released and had much better performance for a fraction of the cost. There's a lot of optimism and hype around DevDay within the community. And yet, I can't help but feel slightly uneasy. I'm worried that most of the 20 features dropping tomorrow will be largely unavailable for the Plus and perhaps even $100 plan users. Pushing the frontier with new powerful tools is great. As long as you don't need an expensive subscription to try it. Especially since the Plus plan is how most new subscribers get exposed to ChatGPT and its wonderful features. A lot is riding on tomorrow. As Anthropic's newest models expose frontier intelligence to free users, I hope OpenAI will offer some products that compete with them tomorrow, proving that their foremost value is AI for everyone. I hope... but I feel somewhat doubtful. And I say this with great respect for you and the entire OpenAI team. The level of transparency and commitment to fixing your mistakes is the top of its class, and you set a level that other companies in the industry can only dream of achieving. I'm looking forward to tomorrow's event and am hopeful that I'll be able to try the latest and greatest stuff the team has been cooking for the past few months. Regards, A long-time ChatGPT subscriber