CodeClawd

Path towards spiritual hacking. Ai enthusiast. Just vibe coding SaaS apps and burning tokens. Building tools and educating young minds how to use Ai to build…

[CodeClawd @codeclawd]

@bcherny even with opus 4.8 selected the Claude Code on web and phone stops mid way. Giving errors

[Suni @suni_code]

Never met a Vibe Coder who could explain this. 😭

[CodeClawd @codeclawd]

@ClaudeDevs After opus 4.5 this was the next Quantum Leap after using Fable fire three days going back to opus felt like rising a bicycle when you were driving a Ferrari

[ClaudeDevs @ClaudeDevs]

As a result of a US government directive, we are suspending access to Claude Fable 5 for all users. You can continue to use all other Claude models.

Here’s what this means for you:

Across Claude products, new sessions will run on your selected default model or Opus 4.8, and existing Fable 5 sessions will end with an error.

On the Claude Platform, requests to Fable 5 will also return an error. Please update your integrations to other Claude models.

We know this is a disruption to your workflows; we appreciate your patience and support.

[CodeClawd @codeclawd]

@AnthropicAI This is such … finally something which was amazing. Hopefully yall can get it back

[AlphaFox @alphafox]

Times are getting tough, everyone is getting replaced by AI: 😞

[CodeClawd @codeclawd]

@bcherny @ClaudeDevs Thank you for this update. in retrospect the opus 4.7 was more of a downgrade from from 4.6. had reverted back to 4.6 for coding only 4.7 for ui components. 10 hours in & the workflows is amazing. unless there is better way of prompting while debugging would be great

[CodeClawd @codeclawd]

Yup. Going back to opus 4.6. Its like talking to chatgpt 5.4 the way opus is talking and the quality has declined

[Jason Akatiff @smaxor]

I just mass-reverted every Claude Code project back to Opus 4.6.

Here's why.

I've been running Opus 4.7 for the past few days across multiple production builds. Lead routing platform. Case Cleanse. Real software that ships to real clients.

And something was off.

Tasks that Opus 4.6 would knock out in 10 to 15 minutes were taking over an hour. Sometimes 90 minutes. Same prompts. Same project structure. Same everything.

It would eventually finish. But the output was worse.

More hallucinated file paths. More circular refactoring that went nowhere. More "let me restructure this for you" when nobody asked for a restructure.

I'm not a casual user. I've pushed billions of tokens through Claude Code building full production platforms. Hundreds of sessions. I know what good output looks like from this tool because I've been living inside it every single day.

So I rolled back to Opus 4.6 with the 1 million token context window.

First task after the revert? Done in 12 minutes. Clean code. No detours.

Here's how to do it yourself if you're running into the same issues.

The quick way is to just type /model inside any Claude Code session and select claude-opus-4-6 from the list.

But if you want it locked in across all your projects, add it to your settings.json:

{ "model": "claude-opus-4-6-1m" }
Important detail here. If you just set it to claude-opus-4-6 you're going to get the default 200k context window. That works fine for smaller tasks but if you're running complex multi-file builds you want the full million tokens. The 1M model string is claude-opus-4-6-1m. That "max" in the model name is what unlocks the extended context. Don't skip it.

The lesson here isn't "new model bad." It's that newer doesn't automatically mean better for your specific workflow. Everyone's chasing the latest model drop like it's a sneaker release. Meanwhile the people actually shipping software are benchmarking against their own output, not a leaderboard.

If you're building with AI every day, you need to treat model selection the same way you'd treat any other infrastructure decision. Test it. Measure it. And don't be afraid to downgrade if the numbers tell you to.

The best model is the one that ships your product.

[CodeClawd @codeclawd]

@arckollect While you were right on the article but the part about Claude Code is wrong. Fear mongering 🧐

[CodeClawd @codeclawd]

@arckollect While you were right on the article but the part about Claude Code is wrong. Fear mongering 🧐

[CodeClawd @codeclawd]

@ClaudeDevs Does it affect the Claude Code tokens usage or that will be reduced as well. Also, what about local harnesses for testing on local apps and No I am not talking about open claw @bcherny @claudeai

[CodeClawd @codeclawd]

@alifcoder Hustle… does it work

[CodeClawd @codeclawd]

@alifcoder Hustle

[CodeClawd @codeclawd]

@alifcoder No response at all

[CodeClawd @codeclawd]

@alifcoder Hustle. Like and followed

[CodeClawd @codeclawd]

@vulturetrades $SPY

[CodeClawd @codeclawd]

@marryevan999 still waiting ....

[CodeClawd @codeclawd]

@marryevan999 done.

[CodeClawd @codeclawd]

@marryevan999 Claude

[CodeClawd @codeclawd]

@jordan_ross_8F waiting

[CodeClawd @codeclawd]

@jordan_ross_8F marketing