./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.

Friday, 2026-10-09·last updated 2026-10-09 23:05 UTC·9/9 feeds ok

[ hardware ]12

• b11539 — llama.cpp
• b11538 — llama.cpp
• b11537 — llama.cpp
• b11535 — llama.cpp
• Qwen 3.8 Flash Next-GSQ-RCO-IQ2_XS at ~21 tok/s on just an RTX 3060 12GB + 16GB DDR4 RAM… — r/LL
• b11534 — llama.cpp
• b11533 — llama.cpp
• GMKtec EVO-X3 Mini-PC Review Looking at a 128GB AMD Local AI Box — STH
• b11532 — llama.cpp
• b11531 — llama.cpp
• b11530 — llama.cpp
• Tested Mellum2.1-12B-A2.5B on PI Coding Agent - surprisingly usable, but not great at on… — r/LL

[ new models ]6

• Just another purely open-weight models benchmark — r/LL
• GLM 5.3 Flash opensource @ the top of Artificial Analysis Cyber Index over Claude. — r/LL
• Impactful scheduling for GPU clusters — HF
• Qwen-Image-2.1-Turbo released! — r/LL
• Stepping away from Benchmarks and Code, what models are you using for Creating writing p… — r/LL
• Running the uncensored Qwen3.8-27B (HauhauCS) on a 4090 at 262K context and ~130 tok/s — r/LL

[ local ai ]12

• H2O-Lightning-4B: Apache-2.0 4B Decision model, official #1 open model on JevBench (abov… — r/LL
• OpenAI's math findings built on stolen user data — r/LL
• A 24KB stand-alone HTML-LLM that can generate consistent stories — r/LL
• Qwen/Qwen-Image-2.1-Turbo · Hugging Face — r/LL
• tencent/Youtu-Parsing-Omni · Hugging Face — r/LL
• lightonai/LightOnOCR-3-4B · Hugging Face — r/LL
• Ahh, it all makes sense now for the 64Gb DGX Spark - [Samsung expects 780% quarterly ope… — r/LL
• [Paper] Stepped MoE: Segment-Level Routing with Configurable Inference Complexity — r/LL
• [Paper] DLoop: Looped Speculative Decoding — r/LL
• [Paper] EngramEdit: Decoupled Knowledge Updates in LLMs through Conditional Memory — r/LL
• LoRA over GGUF: Train Qwen3.8-Flash-Next in 40G VRAM — r/LL
• StepFun Step 5 preview showing up in openrouter — r/LL

[ benchmarks & papers ]2

• Emotional optimization by newsroom AI needs behavioural evaluation — NatMI
• Characterizing the role of reasoning tokens in large language models as state over token… — NatMI

[ tools ]6

• LlamaTale v0.43.0 - MCP Server for story creation — r/LL
• b11536: ui: Models Manager Follow-up Improvements (#30228) — llama.cpp
• GitHub - google-ai-edge/ml-drift: GPU-Accelerated AI/ML Inference — r/LL
• Qwen3.8-27B: 159 tok/s on R9700, 64 tok/s on Strix Halo — r/LL
• Qwen 3.6 35B appreciation post — r/LL
• Port of the TypeScript compiler, checker and lsp to Rust, by LLM — HN

[ healthcare ai ]12

• STAT+: Biotech leader, leading lab leak theorist no longer in consideration for top infe… — STAT
• STAT+: Hospitals sue – again – to halt Trump administration’s 340B rebate pilot — STAT
• STAT+: Pharma’s free lunches for doctors boost inappropriate prescribing, analysis finds — STAT
• STAT+: Medicare Advantage star ratings trigger new bonuses, and uncertainty — STAT
• STAT+: Life Biosciences share glaucoma study updates for its cellular reprogramming drug — STAT
• STAT+: Up and down the ladder: The latest comings and goings — STAT
• How these census changes could threaten public health — STAT
• STAT+: Pharmalittle: We’re reading about a setback for Argenx, Lilly suits over copycat… — STAT
• STAT+: Will MAHA survive the midterms? — STAT
• Opinion: Proposed U.S. census changes could be catastrophic for public health — STAT
• Opinion: 50 years ago, my colleagues and I pursued basic research science that became re… — STAT
• The fall of Roe dramatically changed abortion access. This researcher used Reddit to cap… — STAT