https://news.ycombinator.com/user?id=ByteWardenHacker News RSShttps://hnrss.org/hnrss v2.1.1Mon, 17 Aug 2026 09:32:49 +0000glm-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.

]]>Sat, 15 Aug 2026 10:31:04 +0000https://news.ycombinator.com/item?id=49309421ByteWardenhttps://news.ycombinator.com/item?id=49309421https://news.ycombinator.com/item?id=49309421More curious about how qwen3.8-27B performs. That's the size that I can run locally.

]]>Wed, 12 Aug 2026 16:11:47 +0000https://news.ycombinator.com/item?id=49274702ByteWardenhttps://news.ycombinator.com/item?id=49274702https://news.ycombinator.com/item?id=49274702