4 ms·
I ran a quick eval to see what this looks like qualitatively vs just calling Opus 4.7 or GPT 5.5 directly. As expected, Fusion was 7x slower and 4x the cost.
by michaelbuckbee 5mo ago
I ran a quick eval to see what this looks like qualitatively vs just calling Opus 4.7 or GPT 5.5 directly.
As expected, Fusion was 7x slower and 4x the cost.
This isn't a knock against it, just that it I think this places Fusion into a "use it only when you need it" category.
https://3fpi5avcqq.evvl.io/ https://3fpi5avcqq.evvl.io/
- IanCal 5mo agoWhich models were you using under this? If you used the quality default as exists in the interface, it makes sense that it was ~4x the cost as it'd be 3 frontier models judged by one of those. The idea would be to use fusion with simpler, cheaper models.
- nielsole 5mo agoSounds like fusion would be a really good distillation target?
- galsapir 5mo agoyeah its really counterintuitive i think; i.e, getting the right framework and structure for this to work probably isn't trivial, models really hate playing well together. i wonder how their version would fair in real world use.
- jimmypk 5mo ago[flagged]
- awllau 5mo agoVery cool, thanks for sharing. Speed was my biggest concern and they didn't mention anything about it.