{"author":"cmrdporcupine","children":[{"author":"selectodude","children":[],"created_at":"2026-10-09T14:50:56.000Z","created_at_i":1791557456,"id":50021337,"options":[],"parent_id":50019783,"points":null,"story_id":50000488,"text":"Yeah, they all do it though. And now that models are being natively trained at FP8&#x2F;NVFP4, I\u2019m not sure it matters.<p>Only way you can really know you\u2019re getting the full model is to host it yourself. Every inference provider has every reason to lie and it\u2019s impossible to find out the degree to which they are.","title":null,"type":"comment","url":null}],"created_at":"2026-10-09T12:53:55.000Z","created_at_i":1791550435,"id":50019783,"options":[],"parent_id":50013645,"points":null,"story_id":50000488,"text":"DeepInfra is decent but has a bad history of quantizing models","title":null,"type":"comment","url":null}
