Ignore Benchmarks. The best way to determine if a new model is good is:
1. Remove your AGENT/CLAUDE.md
2. Remove all your skills
3. Test it on work you do everyday and give it the minimal amount of information to complete the task.
4. Judge if it was better, then add skills
@MiaAI_lab@SpaceXAI@cursor_ai@poteto I love how fast Grok 4.6 is. I have noticed it does struggle with certain tasks a lot more then gpt and fable and it is horrible at design. The best thing for me has been it is really good at AutoCAD for some reason.
I have been working on my own version of Whispr Flow. It is open-source. Still validating it for Mac, but it is very nice free alternative. https://t.co/6GoL6IgQZx
Ignore Benchmarks. The best way to determine if a new model is good is:
1. Remove your AGENT/CLAUDE.md
2. Remove all your skills
3. Test it on work you do everyday and give it the minimal amount of information to complete the task.
4. Judge if it was better, then add skills
I have been working on my own version of Whispr Flow. It is open-source. Still validating it for Mac, but it is very nice free alternative. https://t.co/6GoL6IgQZx
@OmriBuilds I’m assuming you made this yourself? A couple things consider using parakeet v2 for English or v3 for multimodal. It is crazy fast. Moonshine is a slower option but requires less memory. Also, if you haven’t build in AI formatting. It makes the app feel like whisprflow
@levelsio I don’t understand how we are not using people’s health data to help inform doctors. Make it easier to treat stuff before it becomes a major issue.