Simon Pantic

[Simon Pantic @Simon_Pantic]

@videvjs 4.8 ! Dis lui de ré-écrire un mauvais skill, ou une doc, normalement la différence devrait se voir immédiatement = moins de mots, plus d'information et reste lisible.

[Simon Pantic @Simon_Pantic]

@thsottiaux Opus 4.8 is really good. I hope you guys are keeping up and will release GPT-5.6 soon, because I only have room for one Max plan subscription.

[Simon Pantic @Simon_Pantic]

@serenaa_ge This is the first benchmark that looks very close to my own experience... I already found that 5.4 was slightly stronger than opus 4.7.

[Simon Pantic @Simon_Pantic]

@thsottiaux Unfortunately, it coincides perfectly with my weekly reset…

[Simon Pantic @Simon_Pantic]

Le vrai problème concerne surtout les gens qui, comme moi, font beaucoup de workflows A2A entre systèmes (Claude Code vers Opencode dans mon cas). Avant, c’était faisable avec un seul abonnement. Aujourd’hui, il faut payer des extras et les modèles Anthropic brûlent les crédits à une vitesse folle. C’est honnêtement ce qui m’a récemment poussé vers Codex, en plus du fait que GPT-5.5 me semble nettement meilleur qu’Opus 4.7 pour mes usages.

[Simon Pantic @Simon_Pantic]

@rezoundous No… Also, I recommend using Low + Fast for everything that involves a very tight see-and-adjust loop (typically frontend design).

[Simon Pantic @Simon_Pantic]

@thsottiaux Adding Images 2.0 and GPT-5.5 Pro would be the next major step. I understand the hesitation around A2A workflows, but a transparent daily usage cap per call would make a lot of sense for these models.

[Simon Pantic @Simon_Pantic]

@pbakaus This is awesome, but why free ? It's a full product here with your own vision I don't see any interest of open sourcing.

[Simon Pantic @Simon_Pantic]

@thdxr I’m actually building something on top and it’s very hard to keep the repo synced 😅

[Simon Pantic @Simon_Pantic]

@brivael Désactive 1M context / adaptative thinking et switch à "max effort" en début de session (seul moyen car ce mode n'est pas atteignable par défaut). Et ça redevient le Opus 4.6 d'avant. Anthropic est sous l'eau en usage donc ils ont trouvé tous les subterfuges possibles.

[Simon Pantic @Simon_Pantic]

@_StanGirard Hum... are you aware of Open Code ? Because it's extremely good and probably what you need here.

[Simon Pantic @Simon_Pantic]

@CameronFoxly love the website and open source also ! congratulations.

[Simon Pantic @Simon_Pantic]

@robinebers Opus 4.5 stands out for one key reason: a very good balance of speed and capability. It explains exactly what it’s doing with incredibly clear, concise outputs and neat ASCII diagrams, a huge improvement over Codex, which is long thinking and cryptic.

[Simon Pantic @Simon_Pantic]

@engelnyst @steipete Nop, just repeating the same output endlessly.

[Simon Pantic @Simon_Pantic]

@pvncher I just finished my first coding session with Opus 4.5, and it was so good that I immediately canceled Codex pro plan.

[Simon Pantic @Simon_Pantic]

@pontusab I just discovered your app. I’m building something similar for the French market. My agentic system is similar, but organized by tools instead of business domains to minimize context and model size for each sub-agent based on task complexity.

[Simon Pantic @Simon_Pantic]

@pvncher The limit might be low, but remember that Codex has to work within a 200k context window. Function calls are common, and if even 10% of that window gets filled with useless output, which happens often during development, it’s already in trouble.