Julien Bernardo

ex–data scientist → founder 🧑‍💻

Building AI apps to “make the world a better place” 🌍

Early GANs researcher (before it was cool) 👠

[Julien Bernardo @JulienFBernardo]

Has anybody switched from Opus 5.5 to Sonnet 5.5 as their daily driver?

What are your thoughts?

[Julien Bernardo @JulienFBernardo]

@blueemi99 Thats only on max effort, which no one should ever use on either model

[Julien Bernardo @JulienFBernardo]

@OpenAI Bot grok?

[Julien Bernardo @JulienFBernardo]

Sonnet 5.5 > Opus 5.5 on terminal bench!

It feels like Opus 5.5 just came out and now its already beaten by a lower tier model?

[Claude @claudeai]

Sonnet 5.5 improves on Sonnet 5 across benchmarks, in some cases dramatically.

It’s a faster, lower-cost complement to Claude Opus 5.5, strongest at well-scoped everyday tasks, fixing bugs, and creating polished documents, slides, and spreadsheets.

[Claude @claudeai]

Sonnet 5.5 improves on Sonnet 5 across benchmarks, in some cases dramatically.

It’s a faster, lower-cost complement to Claude Opus 5.5, strongest at well-scoped everyday tasks, fixing bugs, and creating polished documents, slides, and spreadsheets.

[Julien Bernardo @JulienFBernardo]

@HarryStebbings Their GTM is among the best

[Julien Bernardo @JulienFBernardo]

@sashikantsingh_ Hahah deleting all the tests

[Julien Bernardo @JulienFBernardo]

@melvindvivas Use Astra as a reviewer, its still very good at that. But choose wisely you will only be able to review 1 or 2 PRs before usage is burnt

[Julien Bernardo @JulienFBernardo]

@theo Now do it without fast mode

[Julien Bernardo @JulienFBernardo]

@developedbyed It's not even about having the frontier models, OpenAI needs to fix Sol 6 to actually be comparable to Opus 5.5. Or get Astra to burn through limits less quickly

[Julien Bernardo @JulienFBernardo]

@Da7_Tech It's dirt cheap and token efficient. I agree its not the best for coding, but if you're using it in your own agentic product it's hard to beat. The chinese open models are very unreliable outside of pure coding.

[Julien Bernardo @JulienFBernardo]

@antonioventre_ Dont half the users on IG and FB have sound turned off?

[Julien Bernardo @JulienFBernardo]

@elonmusk You've got the compute (Colossus), data (Cursor), and the brains (SpaceX). Bullish on Grok being the future daily driver for devs that care about cost efficiency and speed! Might take a couple releases Grok 5+

[Julien Bernardo @JulienFBernardo]

@TokenGremlin They are favoring enterprise plans > regular plans

[Julien Bernardo @JulienFBernardo]

@uday_devops Its already happening and im one of them

[Julien Bernardo @JulienFBernardo]

@itsalicesoul Dont let your context window creep up above 400k, its the sweet spot according to @mattpocockuk
Beyond that you lose performance and increase usage, better to go on a new thread

[Julien Bernardo @JulienFBernardo]

@VadimStrizheus OpenAI better blow our minds this coming tuesday DevDay to make up for their weak release

[Julien Bernardo @JulienFBernardo]

With Codex down, and Opus 5.5 on $20 plan doing more work for me than Astra on $100 plan. The OpenAI team better rock the world on tuesday DevDay @thsottiaux

[Julien Bernardo @JulienFBernardo]

The image generation and computer capabilities from GPT models are incomparable. i miss them ever since my switch to opus 5.5, but the extra limits and coding capabilities make up for it. Astra is still my fav because it just gets shit done end to end without me, but the limits :(

[Julien Bernardo @JulienFBernardo]

Sol 6 feels like a terra model! I reach my weekly limit with Astra in a single day on the $100 plan. Switched to Opus 5.5 and even on the $20 plan it takes me 3 days to reach the limit.

Honestly I still prefer Astra, its more thorough and is more autonomous. Closest model to making me feel like I have an employee. But unsusable with current limits :(