Twitxy Saved

Ben Clavié

Verified account Mixedbread

@bclavie

regressing linearly on a daily basis. wife guy who does retrieval. research @mixedbreadai, prev answerdotai - dms work randomly, responses might be very slow
吉祥寺 ben.clavie.eu Joined April 2016
1.5K Following 7K Followers 4.1K Posts

Latest posts 5

Pinned post

Ben Clavié Verified account @bclavie
I'm so excited to introduce this!

We've worked on a million different moving parts to produce this. I'm fairly confident it's the best multimodal model that exists, period -- and it's not too shabby at pushing back the LIMITs of retrieval either...
Mixedbread @mixedbreadai Mar 12
Introducing Mixedbread Wholembed v3, our new SOTA retrieval model across all modalities and 100+ languages.

Wholembed v3 brings best-in-class search to text, audio, images, PDFs, videos...

You can now get the best retrieval performance on your data, no matter its format.
View quoted post on X
39 replies 43 reposts 418 likes 151K views View on X
Ben Clavié Verified account @bclavie
I feel like Neon is one Claude Code-level in-house harness away from feeling like a Sonnet 3.5 monumental step forward in the pareto frontier of "feels good to use, is fast, does all I want correctly"
12 replies 5 reposts 393 likes 12K views View on X
Ben Clavié Verified account @bclavie
6.2 Astra $0.02 vibe check:

- Noticeably better than previous Astra&Sol.
- Not my personal cup of tea at all. It retains the (recent-ish) quirks of GPT models that it's bad at exploration and either refuses to make reasonable assumptions from a prompt, or makes assumptions that make extremely little sense in context, which is where Opus 5.5 shine. Not sure if it's a harness issue or a model issue, however.
- Horrible to steer (again, not sure if codex or model issue)
- Very good implementer to have as a workhorse driven by Opus/Fable. Short tasks and extremely well scope prompts alleviate the above problems.
- Token efficiency is nice, doesn't burn through usage very fast.