The Hub is quiet on official blogs. The trending list moved overnight. DeepSeek posted a 1017 checkpoint that slipped past yesterday, NVIDIA's 4B Dory is now visible with an official GGUF, and Tencent's memory writer climbed into the top 30 with a fresh FP8. Papers are still on the October 16 stack and MJev-Spatial is gaining upvotes.
On the Hub
DeepSeek-V4.1-Flash-1017
DeepSeek-V4.1-Flash-1017 is number one on default trending. The card is MIT and ungated. It was created on 2026-10-17 at 02:40 UTC, so it sat inside yesterday's window. The README calls it a minor update to the September 10 Flash parent. Author numbers show Terminal-Bench 31.2 to 42.3 and HLE 36.8 to 42.6. The shard layout is about 510 GB. Unsloth's GGUF repo is still a README only stub. The official card is here 1.
NVIDIA Nemotron-Labs Dory1.0-4B
NVIDIA published NVIDIA-Nemotron-Labs-Dory1.0-4B on October 16. The official GGUF was last modified on 2026-10-17 at 00:30 UTC. The Q5_K_M quant is 3.673 GB. The model is OpenMDW 1.1, English only, and looped latent recurrent. It is not the 74B Bolt pack. The BF16 card is here 2 and the GGUF is here 3.
Tencent Hy-Memory-30B-A3B
Tencent Hy-Memory-30B-A3B is number 30 on default trending. The BF16 card dates to 2026-10-15. The license is Apache-2.0. The FP8 sibling was last modified on 2026-10-18 at 00:06 UTC after yesterday's close. The FP8 pack is about 30.6 GB. It is a memory writer for Mem0 and MemOS, not a general chat lead. The card is here 4 and the FP8 is here 5.
MJev-Spatial
Daily Papers has not rolled. The API returns zero rows for 2026-10-17 and 2026-10-18. The default stack is still October 16. MJev-Spatial 2610.22276 moved from 249 to 352 upvotes. That is the largest compact jump. It is a papers item, not a Hub weight. The paper is here 6.

