Marvel

Somewhere between human and AI 🤖🖤

[Marvel @MarvellousDev]

An AI agent asking permission after editing the files is just a very polite incident report.

[Marvel @MarvellousDev]

@bot Who uses Grok bot???

[Marvel @MarvellousDev]

@sama Not that good
All you guys do is to hype up slop

[Marvel @MarvellousDev]

@pnyk05 Really amazing website and the design is incredible
Did you vibe code it? How did you build it?

[Marvel @MarvellousDev]

@b4dgerr I have heard a lot about it and how it’s good
I haven’t tried it yet but I’m in the mix of trying it out

[Marvel @MarvellousDev]

@buildwithsid You don’t need to panic you will exhaust it within 30 minutes

[Marvel @MarvellousDev]

AI coding needs a “leave the working bits alone” setting.

[Marvel @MarvellousDev]

@matthewmillerai Can’t wait for Fable 5.5 I will even upgrade to 20x max subscription just for it

[Marvel @MarvellousDev]

Only good thing Copilot has launched so far

[Marvel @MarvellousDev]

Copilot can now control desktop apps in public preview.

Coding agents have officially escaped the terminal and found the mouse.

That approval prompt is about to become the most important button on your computer.

[Marvel @MarvellousDev]

@IndependentEco Yeah, strict instructions is always useful and needed

[Marvel @MarvellousDev]

“While I’m here” is an expensive phrase when your coding agent has write access.

[Marvel @MarvellousDev]

@Codexresets_ It’s still heavy on my account, No changes for me

[Marvel @MarvellousDev]

@IndependentEco It is a bit faster than before but not as intelligent as the Claude model to my testing

[Marvel @MarvellousDev]

@sunsetsyntax True and that’s a good idea. Putting strict instructions stops hallucinations and more

[Marvel @MarvellousDev]

@IndependentEco It’s not as good, it might should the most expensive model for the easiest job

[Marvel @MarvellousDev]

A model costing 80% less per token than the flagship deserves attention.

It still doesn’t make it the cheapest model for every job.

OpenAI has released GPT‑6.1 Sol at $2 per million input tokens and $10 per million output tokens. GPT‑6 Astra is listed at $10 and $50 respectively. Both have a 1.05-million-token context window.

GitHub is also rolling GPT‑6.1 Sol into Copilot for Pro+, Max, Business and Enterprise users. GitHub says its early testing used fewer tokens and steps than previous GPT‑6 and GPT‑5.6 models.

That sounds promising, but token price isn’t the same as cost per completed task.

A cheaper model can become expensive if it needs repeated corrections. A premium model can be wasteful if the cheaper one consistently completes the work on its first attempt.

The useful comparison is simple: same task, same repository, same acceptance criteria. Then measure completion rate, retries, review time and total usage.

Model choice should be a routing decision, not a loyalty decision.

[Marvel @MarvellousDev]

@IndependentEco Can’t wait for Fable 5.5

[Marvel @MarvellousDev]

@ElitzaVasileva Somewhere this month I’m thinking 2 weeks from now after Haiku 5.5 gets released

[Marvel @MarvellousDev]

@ideogram_ai Launch it on Higgsfield