./daily — ← back to field notes

Daily links

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.

Monday, 2026-10-19·last updated 2026-10-19 16:12 UTC·9/9 feeds ok

[ hardware ]10

• b11734 — llama.cpp
• b11732 — llama.cpp
• b11731 — llama.cpp
• DeepSeek V4.1 Flash at 2.61 bpw + Engram: 26 t/s on dual RTX 3090s with dsx — r/LL
• b11729 — llama.cpp
• b11727 — llama.cpp
• b11730: cuda: cap MMQ Stream-K blocks at available work (#30553) — llama.cpp
• b11724 — llama.cpp
• GL.iNet Comet Q GL-RMQ1 Review A Single USB Type-C Cable Remote KVM Device with Tailscal… — STH
• Embedding model can be used to search text inside image — r/LL

[ new models ]10

• inclusionAI/Ming-3.0-audio-tiny · Hugging Face — r/LL
• inclusionAI/Ming-3.0-audio-flash · Hugging Face — r/LL
• b11737 — llama.cpp
• b11736 — llama.cpp
• b11735: metal : fix moe_reduce fusion race when dst aliases its inputs (#30795) — llama.cpp
• Don't sleep on Nemotron 3.5 Super VL — r/LL
• Nemotron 3.5 Bolt is okayish for small multiuser setups — r/LL
• Little Sweet Room Challenge | Strata Fork | Qwen 3.8 FN NVFP4 | DeepSeek Harness — r/LL
• 72 hours, four big open weights drops. The only one that fits in 24GB came from NVIDIA… — r/LL
• Scaling Indic Pretraining: 2.8 Trillion Tokens and an Open Data Generation Pipeline — HF

[ local ai ]12

• ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-Coder-GGUF updated — r/LL
• inclusionAI releases Ming-3.0-audio-flash: 126B MoE / 6.5B active, audio understanding +… — r/LL
• Minimax M3.1 Flash is 428B 23a — r/LL
• A bare IPython harness ties production coding agents with Qwen3.5-9B on TB-Lite, and too… — r/LL
• MiniMax M3.1 Flash released, Model weights Oct 23 — r/LL
• llama server's default port changed from 8080 to 9931 — r/LL
• MiniMax-M3.1-Flash is coming — r/LL
• AI labs report benchmark scores 6.4% higher than independent testers, on average — r/LL
• incoai/Qwen3.8-Flash-Next-DFlash2 — r/LL
• Kapitsa: open-source Apple Silicon optimized Metal engine built for Qwen and Nemotron. S… — r/LL
• DeepSeek-V4.1-Flash-1017 on AA & Vals index — r/LL
• TempoPFN: Synthetic Pre-training of Linear RNNs for Zero-shot Time Series Forecasting — r/LL

[ benchmarks & papers ]5

• Qwen 3.8 Flash Next optimized for two CMP 170hx GPUs — r/LL
• Towards understanding and monitorability of reasoning LLMs — NatMI
• Reusability report: multi-sample omics network inference operates in distinct robustness… — NatMI
• If you don't run your own LLM proxy, how do you know whats going on? — HN
• Hybrid Reasoning Agents: LLM agents that think in latent space, act through text — HN

[ tools ]3

• Harness Engineering 101: The LLM is the brain, the harness is the body — HN
• Calling all GPUs: help us benchmark SonicBoom v0.0.3 — a compiler-first AI runtime that… — r/LL
• Google Open Sources Harness2 for Self Improving AI Agents — r/LL

[ healthcare ai ]12

• STAT+: The umpires getting rich off surprise medical bills, and MLB job lock — STAT
• STAT+: China moves to restrict foreign biotech acquisitions — STAT
• Two former Republican lawmakers say the GOP’s vaccine position is a liability — STAT
• STAT+: Pharmalittle: We’re reading about Roche investments in Germany, a Gilead bowel di… — STAT
• STAT+: Gilead’s IBD drug eliminated symptoms for patients, mid-stage study finds — STAT
• STAT+: Top immunology association divided over $8.4-billion California ballot measure to… — STAT
• STAT+: Does Isomorphic Labs have what it takes to revolutionize drugmaking? — STAT
• Opinion: My mother killed herself. I wish assisted suicide had been an option — STAT
• Opinion: Measles is back — and becoming a midterm liability for Republicans — STAT
• STAT+: How the No Surprises Act’s nonprofit arbitrators are operating like ‘mini gods’ a… — STAT
• STAT+: Cambridge scientist David Sabatini speaks out after courtroom victory: ‘The harde… — STAT
• Healthcare AI needs a different model-development flywheel — HN-HC