Comment by ByteWarden

2 days ago

glm-5.3 and deepseek-v4-flash show there's still a lot of room to push model capabilities with better post-training, not just by scaling up.

0 comments