Error: Agent terminated early due to an API error: API Error: Output blocked by content filtering policy (error type unknown, request id ..., model sent to the API: claude-opus-5-5)
18
A big part of the supposed nerfing of models is likely just developers getting used to how good the model is, and no longer being impressed by it. I still have jaw-dropping experiences with Opus 5.5. Without a doubt my favorite model ever.
7
Olav retweeted
point your agent at this and build whatever github.com/anomalyco/rune
Why does Claude Desktop randomly switch from Opus 5.5 to Opus 5? It has happened several times without me noticing. Really annoying.
9
I wonder if this is my last annual subscription to @jetbrains. My first license was PyCharm in 2013, later the All Products Pack. I want to support them, but I barely touch my IDE these days, yet producing heaps more code. What a change vs 2 years ago!
14
I'm getting so much more done with the $100 sub of Claude than the $100 ChatGPT sub, even counting all the resets we got in the last month. I doubt the 2x-ing of limits would have changed that much.
14
Someone should make absurd cooking show videos with real clips combined with AI clips. I would watch that.
7
If you don't have an API, CLI or MCP I will just let my agent hammer your site with computer use. Your choice. 🤡
12
I wish the @deepseek_ai platform had configurable limits to avoid runaway agents. Like $X/24h. I guess I could create an Opencode / Pi extension.
17
Waiting for slow cloud subscription models when you're used to local models at ~200TPS is brutal. Of course it's worth it for SOTA models for many tasks, but even Luna is unbearably slow (and dumb).
3
I wanted a better way to manage floating windows in #niri / #noctalia. Meet stashpad! github.com/oal/stashpad Fully built by Qwen 3.8 Flash Next running on Strata.
54
With the coming abundance of AI created/assisted apps, games, music and movies, we need curators to filter through everything. Not necessarily human curators, but something to filter the slop from the genuinely good stuff.
10
I can't remember GPT 6 Luna ever being this useless before. I gave it a simple task of upgrading Strata. It got the exact version number and the exact path to my local model GGUFs. Yet it did pretty much EVERYTHING wrong. @OpenAIDevs
