./daily — ← back to field notes

The link sheet

One-line pointers to today's local-AI, model, hardware, and healthcare-AI news. No commentary — just the headline, the link, and where it came from.

Sunday, 2026-09-27·last updated 2026-09-27 21:58 EDT·9/9 feeds ok

[ hardware ]12

• Qwen3.8-Flash-Next 177B NVFP4(119GiB): SSD streaming at 9-10 tok/s on one 16 GB RTX 5060… — r/LL
• b11222 — llama.cpp
• Updated from 3x3090(2x3090, 1x3090TI) to 2x5090 — r/LL
• ASUS AI Tech 2026 Seoul South Korea — STH
• b11221 — llama.cpp
• b11218 — llama.cpp
• b11217 — llama.cpp
• b11216 — llama.cpp
• b11215 — llama.cpp
• b11214 — llama.cpp
• b11213 — llama.cpp
• b11212 — llama.cpp

[ new models ]10

• Qwen company already rushed out a Jev competitor. No open weights yet. — r/LL
• b11223 — llama.cpp
• Swift 1.5 Qwen3.8 27b (A must-have for low thinking!) — r/LL
• Imbalanced VRAM usage between two GPUs in llama.cpp. Anyone successfully solve this? — r/LL
• Don't trust frontier models when asking about budget hardware! — r/LL
• Worth going from Qwen3.8 27B to flash next or maaybe deepseek v4 flash? — r/LL
• Nonobench v1.2: 43 LLMs on nonogram puzzles. Open-weight DeepSeek V4 Pro ties for 4th, a… — r/LL
• Which of the 16gb VRAM qwen3.8 27b’s is the best? — r/LL
• Qwen, where's the small stuff? (1B/2B/4B) — r/LL
• Splash fork optimised for M5 Max: ~1.5× faster (1.25× single request) — r/LL

[ local ai ]11

• FT: Corporate America rejects overpriced frontier, embraces open models — r/LL
• Qwen plays World of Warcraft — r/LL
• ... so, yeah. — r/LL
• Naive-N0.5-Flash - 309B-A15.5B — r/LL
• Mimo v2.6 flash MOPD — r/LL
• is switching from llama cpp to vllm worth it — r/LL
• Adding logit penalty for "wait", "maybe" and "perhaps" to Qwen models improves their acc… — r/LL
• How do you guys give your models web browsing capabilities? — r/LL
• The Opus 5.5 posts about motion graphics are cool but Qwen 27B made this on a 4090 — r/LL
• Another "Harness matters" post (codex cli > pi and opencode) — r/LL
• SupersonicLabs/Julia-1 · Hugging Face — r/LL

[ benchmarks & papers ]2

• "As a Language Model": Chat Template Switches LLM Self-Referential Voice — HN
• Guiding large language models to predict edit sequences for molecular synthesizability o… — NatMI

[ tools ]2

• Public MCP server for Canadian privacy law data (free, no auth) - works with any client… — r/LL
• Zer0Fit - Zero-shot predictions, classifications, and regressions using Google ML resear… — r/LL

[ healthcare ai ]3

• Opinion: I worked on the CDC investigation of the anthrax letters. 25 years later, are w… — STAT
• Healthcare AI's next test is integration — HN-HC
• Inside The Medical AI Factory — HN-MED