Saad Ali

Linux/BSD SysAdmin/DevOps/SRE

[Saad Ali @SaadKnight]

@thomheinrich @thsottiaux @onusoz Its addicting. I was able to move away from Claude Code easily.

[Saad Ali @SaadKnight]

@thomheinrich @thsottiaux @onusoz I use 5.6 Sol with Pi agent. Its the best combo I have experienced so far. I tried 5.6 Sol in Codex nad OpenCode, but I could not see agent execution the way I was ised to it with Claude Code. But with Pi you add a plugin and it shows the agent orchestration & activity nicely.

[Saad Ali @SaadKnight]

@bridgemindai I see the reset. It is there.

[Saad Ali @SaadKnight]

@MikeGropp @TheAhmadOsman I have been saying that they are serving quantized versions of their models. They don't have capacity, unless they provide the service to the government.

[Saad Ali @SaadKnight]

@bridgemindai For the government, they make sure that they have enough capacity because the government payments are garunteed. Everyone else are pesants.

[Saad Ali @SaadKnight]

@ClaudeDevs No wonder your models are lazy and stupid these days. If the capacity is tight you must be quantizing your models to fit more and more users in the same capacity.

[Saad Ali @SaadKnight]

@bridgemindai May be they are running a quantized version of the model for the subscription users.

Personally, I haven't checked if the API version is behaving the same or not.

[Saad Ali @SaadKnight]

@WordPress Wordpress always brings vulnerabilities and is the biggest source of compromises on this planet!

[Saad Ali @SaadKnight]

@digitalocean Are you joking? We can't get on-demand GPUs and you are promising spot GPUs 🤣

[Saad Ali @SaadKnight]

@yume_arasaki Thank you so much for the info!

[Saad Ali @SaadKnight]

@yume_arasaki Memory meaning DDR4 RAM

[Saad Ali @SaadKnight]

@yume_arasaki RTX 3090 with 32 GB of memory and RTX 3090 with 128 GB of memory.

[Saad Ali @SaadKnight]

@jun_song Only ridiculously dumb? It feels like it has always been this way! Giving out dumb models to subscription holders is not cool.

[Saad Ali @SaadKnight]

@bridgemindai @AnthropicAI is in the business of selling nerfed models. They want people to cry out and use API tokens to get more money.

[Saad Ali @SaadKnight]

@trq212 Thank you GPT and Grok!

[Saad Ali @SaadKnight]

@claudeai Why? Are you afraid of GPT 5.6? Fable 5 is slow. Dead slow. It's aggressive making people think to switch to other models.

[Saad Ali @SaadKnight]

Anthropic has successfully made me consider switching to OpenAI Codex by rendering its models from being frontier to being inferior.

#Fable5 #ClaudeAI #Anthropic #claudecode #AI #OpenAI #codex

[Saad Ali @SaadKnight]

2/2
Now it feels like they took that racehorse, tied bricks to its legs, and somehow turned it into a donkey.

Slow. Stubborn. Over-filtered. Waiting to be dragged through every step.

Safety matters, but this hurts.
Please don’t nerf the magic out of frontier AI.
#LLM #AICoding

[Saad Ali @SaadKnight]

1/2
Fable 5 before the ban felt like a big magical brain.

You gave it a hint and it ran.
It understood the idea behind the prompt, filled in the gaps, connected the dots, and built things before you even finished explaining them.

It felt alive.
#Fable5 #ClaudeAI #Anthropic #AI

[Saad Ali @SaadKnight]

Hey @AnthropicAI. You nerfed #Claude into being dumber AND somehow made it slower?? I could finish a 45-min documentary while it tweaks a tiny Terraform module. This is not the assistant I signed up for. 🐢

#Anthropic #Terraform #SRE #DeveloperExperience #ClaudeAI