1. If I open several documents in tabs, they always scroll back to the top when I switch between them. It's impossible to work with the documents.
2. Codex still cannot ask interactive questions outside of Plan Mode - this is generally something you should invest more time in. Codex should not only return simple text suggestions in the form of a, b, c, but also solution examples, e.g., in the form of diagrams, graphic visualizations, simple mockups - that would be a real game changer.
1. Allow opening the terminal in tabs, not just at the bottom
2. Improve state retention for open documents -> currently, switching between documents resets their state, and for documents with many pages, you have to scroll to the target location every time
3. Add the ability to edit and comment on Markdown documents -> this would be super useful for content work
4. Add support for vertical tabs in Chrome, currently it doesn't work well
Is GPT‑5.5 really much better than GPT‑5.3/5.4?
My take: sometimes yes, often the gain comes from better harness, not the model itself.
Prompt quality, tool wiring, context packing, and eval discipline can beat a model upgrade.
Run blind A/B on real tasks and compare cost per successful outcome — not hype.
Ask questions in an interactive way, visualize solutions interactively (e.g., mockups, frontend elements, architecture proposals, specifications, plans, etc.). The point is not to just receive text and markdown and not to keep typing 1a, 2b, 3c -> let it work outside of planning mode as well.
1) The ability to open Terminal on the side instead of at the bottom
2) Making "ask mode" available outside of planning mode — ideally a toggle in settings. It's a bit frustrating having to type "1a, 2b, 3c" instead of just clicking an option. An agent that could continuously ask for clarification would seriously improve response quality. Some of us rely on custom SKILLS for planning, and the current UX gets in the way.
Does anyone know — if you link a ChatGPT subscription to OpenCode and use GPT-5.5 Pro there, does it count against your Codex message caps (included in subscription) or is it billed separately? And why is GPT-5.5 Pro available in OpenCode but not in native Codex CLI? @opencode
Really impressed with how Codex is evolving. There are just two things I'm still missing:
1) The ability to open Terminal on the side instead of at the bottom
2) Making "ask mode" available outside of planning mode — ideally a toggle in settings. It's a bit frustrating having to type "1a, 2b, 3c" instead of just clicking an option. An agent that could continuously ask for clarification would seriously improve response quality. Some of us rely on custom SKILLS for planning, and the current UX gets in the way.
I'd really appreciate seeing these features considered 🙏
@thsottiaux I'm starting to think "1a, 2b, 3c" is my new programming language. Still waiting for the day I can just click a checkbox instead of typing algebraic expressions in my chat. Help 😅
#codex #openai
DeepSeek V4 Pro — good for coding, GPT-5.5 agrees 90% of the time that the code it produced is solid.
DeepSeek V4 Flash — fast and really good for simpler tasks and automation.
These are genuinely impressive models. This is exactly what I've been waiting for from open source. 🔥
#deepseek
It's incredible how we're being taken for a ride. I just switched over to a ChatGPT Plus account for a moment. I specifically switched to 5.3-Codex because I wanted to see what the token consumption looks like compared to what it was 2 months ago.
While back then I was able to work on such an account for 2 days for 8-12 hours, now after 3 messages (just a simple 'hi, how's it going' - without any code changes) I already have 30% less of the daily limit...
This isn't cool...
#openai #codex #chatgpt