deepseek-v4.1IMAGE
DeepSeek V4.1 Flash beats Meta's 816B open-weight model on coding benchmarks
The post says V4.1-Flash-1017, with 552B total and 16B active parameters, outperforms Meta's 816B open-weight model released two days earlier on reported coding benchmarks at a fraction of the price.
Sahil Panhotra | Indie Builder
@SahilPanhotraArchived 9 days agoOriginally posted 2026.10.18
Sahil Panhotra | Indie Builder @SahilPanhotraOriginally posted 2026.10.18
meta released an 816B open-weight model just 2 days later deepseek dropped V4.1 Flash 1017 with 552B parameters and somehow managed to outperform it on reported coding benchmarks at a fraction of the price and they're doing this with just 16B active parameters for output how the fuck is deepseek managing to do this 😭
This post quotes another post: x.com/sahilpanhotra/status/2111805025077043376
x.com/sahilpanhotra/status/2111847921088110776Archived 2026.10.19 12:04