./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.

Thursday, 2026-10-01·last updated 2026-10-01 01:34 UTC·9/9 feeds ok

[ hardware ]12

• b11308 — llama.cpp
• b11307 — llama.cpp
• b11306 — llama.cpp
• Least to most expensive (Somewhat modern) GPU's with 32gb of vram (Under $1600) Based on… — r/LL
• b11304 — llama.cpp
• b11303 — llama.cpp
• b11302 — llama.cpp
• b11301 — llama.cpp
• b11299 — llama.cpp
• Gigabyte TRX50 AERO D Motherboard Review — STH
• b11298 — llama.cpp
• b11297 — llama.cpp

[ new models ]5

• If one hour of AI is costing me 0.12€ is paying for frontier a cheaper option? — r/LL
• Qwen3.8-27B-pi: Effort-Ordered Reasoning for Agentic Coding — r/LL
• You can now run Qwen 3.8 27B on AMD NPUs via FastFlowLM (at a killer 1 tps decode) — r/LL
• BAAI/AREX-2 - 27B - Agent model based on Qwen3.8 27B — r/LL
• NVIDIA Kumo Tabular Sets a New Accuracy-Efficiency Frontier for Tabular Prediction — HF

[ local ai ]12

• AI CEO Interviews (2026) — r/LL
• Two open-weights releases: Victoria (Qwen3.8-Flash-Next with 44% of experts cut, 70% Ter… — r/LL
• Index-Translate: 150 text languages, plus document translation, multilingual subtitles a… — r/LL
• What's going on with DGX Spark? Price up $2k in 1 week? — r/LL
• Another Ling model comes out, same receipt, 2 weeks free to use, then open source. Chine… — r/LL
• What was the mafia meeting of the tech criminals about at the White House? — r/LL
• i would like to learn deeply about fine-tuning local models before burning money — r/LL
• Update: Yandex/AliceAI 80B-A3B fine tune progress — r/LL
• We just open-sourced the world's fastest WebGPU kernels for local AI on Hugging Face — r/LL
• add GLM-5.3-Flash (GLM5-Next) support (#27773) · ggml-org/llama.cpp@649dcb1 — r/LL
• Oído: speech recognition that beats Whisper-tiny, running on a $5 microcontroller (open… — r/LL
• add GLM-5.3-Flash (GLM5-Next) support by timkhronos · Pull Request #27773 · ggml-org/lla… — r/LL

[ benchmarks & papers ]3

• Open TTS Leaderboard: Scalable Evaluation for Multilingual Text-to-Speech and Voice Clon… — HF
• Towards foundation-style models for energy-frontier heterogeneous neutrino detectors via… — NatMI
• MicroLLM Lab – Try 7 tiny LLM's in the browser — HN

[ tools ]2

• Update: Strix Halo + R9700 with llama-halo-hybrid - now beats DGX Spark — r/LL
• Best approach for automatically tagging local music collection? — r/LL

[ healthcare ai ]12

• STAT+: Trump’s watered-down Medicare drug pricing rule saves 96% less than initial propo… — STAT
• STAT+: UnitedHealth investor lawsuit moves forward on allegations it ‘manufactured earni… — STAT
• STAT+: HHS announces new efforts to speed up, expand clinical trials with AI — STAT
• Healthcare AI needs a different model-development flywheel — HN-HC
• U.S. poised to boot legal immigrants from Medicaid, including refugees and sex-trafficki… — STAT
• STAT+: What health tech leaders are talking about in Washington policy circles — STAT
• STAT+: Eli Lilly’s Zepbound combined with amylin drug led to significant weight loss in… — STAT
• Rural MAHA followers say Trump hasn’t improved health of their communities, poll finds — STAT
• STAT+: Three biotech updates, including accusations of theft and and a new lobbying lead… — STAT
• STAT+: Pharmalittle: We’re reading about Beam suing over IP theft, a Trump rule on impor… — STAT
• CDC measles death count still lags Pennsylvania’s — STAT
• Opinion: America’s biotech race with China must be won at home — STAT