Topic
Self-Hosting & Hardware
Self-hosting is the control baseline every alternative gets measured against: which models fit which GPUs, what the ops burden really costs at scale, and how to harden an endpoint so it isn't one of the hundreds found open on the internet.
Start here: Self-Hosting vs Inference APIs in 2026: The Real Cost Math — DeAI's reference piece on this topic.
Security feeds flag a botnet living inside self-hosted AI stacks
Security X accounts are amplifying Lumen's PoeLLM findings: 3,400+ servers running LiteLLM and Ollama hijacked since April, with C2 hidden in a GitHub poem.
This Week in DeAI: The Open-Weights Supply Wave Gets a Date
Mistral, Reflection, and Aleph Alpha all promised weights this October; Ecosia already switched, and the deprecation wave shows why owning the file matters.
DeAI Daily Brief — 6 Oct 2026
Today in DeAI: Reflection AI launches the 501B Beam, OpenAI ships EU text watermarking with published detection limits, Iterate.ai launches TEE inference.
Google, OpenAI, Azure, Snowflake retire hosted models in mid-October
Google retires 16 open-model endpoints off Vertex on Oct 21 and OpenAI shuts 12 legacy GPT snapshots Oct 23; Azure and Snowflake cull models the same week.
DeAI Daily Brief — 4 October 2026
Today in DeAI: Aleph Alpha open-sources the 78B Kolibri-1 under Apache 2.0, Prime Intellect launches GB200 inference, and a Nesa Chain exploit mints $55M.
Aleph Alpha open-sources Kolibri-1, a 78B sovereign EU model
Aleph Alpha released Kolibri-1, a 78B English-German open-weight model under Apache 2.0, with 3.46B active parameters per token and a 1M-token context.
DeAI Daily Brief — 3 October 2026
Today in DeAI: Underdog launches fully on-device with a16z backing, Ethereum ships zkAPI private inference payments, and NVIDIA adds a $4,999 DGX Spark.
Run GLM-5.3-Flash Locally with llama.cpp (2026): VRAM, Quants, Setup
Z.ai's 320B GLM-5.3-Flash now runs in llama.cpp as of the Sept 30 merge. GGUF sizes from 92 GB, quant quality data, setup commands, and known limits.
NaiveAI ships 309B open-weight MoE with a 1M context window
X is amplifying NaiveAI's Naive-N0.5-Flash, a 309B MIT-licensed MoE with 1M context. The weights are on Hugging Face; the speed claims remain unverified.
DeAI Daily Brief — 23 September 2026
Today in DeAI: 36,769 exposed self-hosted AI endpoints, a 328 GB open-weight security model, and Anthropic's GTG-16008 claim against Xiaomi.
36,769 self-hosted AI endpoints are reachable online — 2.02% gated
A scan counted 36,769 self-hosted AI endpoints online, and only 741 (2.02%) asked for authentication. Self-hosted privacy means nothing without a perimeter.
Laya open-weights a Jev-style decision model, X argues about it
An indie dev open-weighted Laya, a 421M Apache 2.0 rival to TypeSafe's closed Jev API, and X front-page velocity met HN counter-testing that found real gaps.
Xiaomi open-sources MiMo-V2.6-Pro and Flash under MIT
Xiaomi released the MiMo-V2.6 open-weight series on Sept 22: a 524B omnimodal flagship and a 159B Flash model, MIT-licensed, with the RL training code public.
The 6 DeAI stories that mattered this week
The week in open and decentralized AI: EU enforcement goes live, Cohere encrypts hosted inference, Anthropic promises ZDR privacy, and Gemini breaks out.
How to Run Qwen3.8 via API in 2026 (Hosts Compared)
Qwen3.8 ships a self-hostable 27B dense model plus a 2.4T Max flagship with fresh open weights. Run it locally, via API, or both — without guessing.
152,137 exposed Ollama servers found in year-long scan
A year-long internet scan counted 152,137 publicly reachable Ollama servers on port 11434, and barely 3% patched known flaws. Self-hosters must act.
From Ollama to a Private Endpoint: Keep Privacy, Drop Ops (2026)
Move from self-hosted Ollama to a hosted private LLM endpoint with one config change — plus the security, zero-retention, and provider checks that matter.
Self-Hosting vs Inference APIs in 2026: The Real Cost Math
Is self-hosting an LLM cheaper than an API? It hinges on one number: your break-even token volume. Here is the cost framework and a worked example.
