Following opens with early access — sign up for early access to register your interest; it does not switch following on today. Everything you follow will land on one page, in your account menu.

1. Open Source & Self-Hostable

   Ant Group releases the weights of its 560B-parameter Ling-3.1-flash model

   Another very large open-weight MoE model is now self-hostable, and the four-GPU mixed-precision setup brings it within reach of a single server.

   TechNode·2026-10-16
2. Open Source & Self-Hostable

   NVIDIA's Nemotron 3.5 Super, a 120B open vision-language model, is now runnable in Ollama

   A 24 GB-class agentic model with a 1M-token context is a practical option for teams that need to keep agent workloads on their own hardware.

   Ollama / NVIDIA·2026-10-15
3. Open Source & Self-Hostable

   Open-weight models now carry 71.5% of Vercel AI Gateway tokens and 23% of spend

   This is real production routing data, not a benchmark. Teams are switching labs often and sending most volume to cheaper open models, while frontier models keep a large share of spend.

   Vercel·2026-10-15
4. Open Source & Self-Hostable

   Ollama 0.40 runs models on Apple's MLX by default on Apple Silicon

   The runtime under existing Mac setups changes on upgrade, so re-check speed and memory use for the models you already run.

   Ollama·2026-10-06
5. Open Source & Self-Hostable

   Reflection debuts Beam, an open-weight model aimed at enterprises and sovereign AI

   A large US open-weight model competing directly with Chinese open models gives teams that need self-hosted weights another serious option to evaluate, though the compute-efficiency claim is Reflection's own.

   TechCrunch·2026-10-05
6. Open Source & Self-Hostable

   NVIDIA adds a 64GB DGX Spark as the 128GB model's price jumps to $6,950

   Local-inference hardware just got a cheaper entry point and a more expensive top tier, so anyone budgeting for desk-side AI boxes should re-price now.

   ServeTheHome·2026-10-03
7. Open Source & Self-Hostable

   Moonshot's Kimi K3 enters OpenAI's enterprise Codex channel and billing

   Enterprises can use a leading Chinese open model inside an existing OpenAI contract without a new supplier agreement, which changes vendor-risk reviews.

   TechNode·2026-09-30
8. Open Source & Self-Hostable

   Ant Group's Ling 3.1 Flash, a 560B-parameter MoE model, launched with a free trial and promised open weights

   Another large Chinese MoE you can trial free now, and possibly self-host later if the open-weights promise is kept.

   Vercel·2026-09-30
9. Open Source & Self-Hostable

   Ollama adds decision models that return choices and probabilities instead of text

   Cheap, local single-pass classifiers can replace many small LLM calls in an agent, but check the independent evaluations before trusting them for routing.

   Ollama·2026-09-29
10. Open Source & Self-Hostable

    llama.cpp reaches v0.4.0 with new model support and per-slot context limits

    Per-slot context limits and on-demand tensor reading are the practical knobs for serving several users off one local box without running out of memory.

    ggml-org / llama.cpp·2026-09-04
11. Open Source & Self-Hostable

    Ollama 0.33.3 lands, adding image and audio input for gemma4

    Multimodal input on a local runner means image and audio prototypes no longer have to leave your machine.

    Ollama·2026-09-03
12. Open Source & Self-Hostable✓ verified

    NVIDIA opened a proxy that lets one application speak both the OpenAI and Anthropic APIs

    Changing model or provider is normally an application rewrite; a translating proxy turns it into a routing rule you can measure both sides of.

    NVIDIA·2026-09-02
13. Open Source & Self-Hostable

    Hugging Face published 200-plus WebGPU kernels so models can run in the browser

    Inference in the browser is the cheapest deployment there is — no server and no per-token bill — and fast GPU operations across mismatched devices have been the missing floor under it.

    Hugging Face·2026-09-01
14. Open Source & Self-Hostable

    Tencent open-sourced Hy4 preview - 770B total parameters, 49B active, a 1M-token context

    A 49B-active open-weight model with a million-token window makes long-document work a self-hosting decision rather than a closed-API one.

    Tencent·2026-08-28
15. Open Source & Self-Hostable✓ verified

    Alibaba released Qwen3.8-Flash-Next, a sparse open-weight model that activates 6B parameters per token

    A 6B active-parameter model with a million-token ceiling is the shape that makes long-context agent work affordable on hardware you already own, and at a twelfth of flagship pricing it resets what a routine call should cost.

    Qwen (Alibaba)·2026-08-26
16. Open Source & Self-Hostable

    Z.ai shipped GLM-5.3-Flash, a 321B open-weight model built for cheap coding work

    Two large open-weight coding models landed on the same day at flash-tier prices; the cheap end of the coding-agent stack is now an open-weights question, not a frontier-lab one.

    Z.ai·2026-08-26
17. Open Source & Self-Hostable

    Ollama's 0.33 preview lets you switch local models on and off inside Claude's desktop app

    One misplaced system message was busting the cache on every request — worth checking how your own prompt is assembled before blaming the model for being slow.

    Ollama·2026-08-21
18. Open Source & Self-Hostable

    Ollama 0.33 makes a cancelled prefill resumable, and adds a Claude Desktop path

    If you self-host, a broken prefill restore is invisible until you look at latency; this is the release that stops a near-complete cache hit from re-running the whole prompt.

    Ollama·2026-08-21
19. Open Source & Self-Hostable

    llama.cpp starts a stable release channel with v0.2.0

    Packagers and anyone shipping llama.cpp inside a product finally have a version to pin that does not move with every commit.

    llama.cpp·2026-08-21
20. Open Source & Self-Hostable

    Ollama halves time-to-first-token by caching resolved model metadata

    Time-to-first-token is what a local model feels like, so halving it changes whether self-hosting is pleasant enough to use for real work.

    Ollama·2026-08-19
21. Open Source & Self-Hostable✓ verified

    Modular opens the Mojo compiler under Apache 2.0, a week after Mojo 1.0

    A closed compiler is a single point of vendor failure for anything built on it; Apache 2.0 on the compiler is what moves Mojo from an interesting runtime to something a team can commit a GPU codebase to.

    Modular·2026-08-18
22. Open Source & Self-Hostable✓ verified

    Qwen 3.8 27B scores 52 on the Artificial Analysis Intelligence Index, level with models far larger

    A 27B open-weights model at frontier-index parity is the size that actually fits on hardware you own — the point where running it yourself stops being a downgrade.

    Artificial Analysis·2026-08-17
23. Open Source & Self-Hostable

    FreeToken serves a 753B open-weight model from a single workstation GPU

    The gap between "open weights exist" and "I can run them" is a bandwidth-scheduling problem, and this is the clearest published attempt to close it on consumer hardware.

    arXiv·2026-08-17
24. Open Source & Self-Hostable

    llama.cpp folds --mmap, --no-mmap, --mlock and --direct-io into one --load-mode flag

    Anyone pinning llama.cpp in a Dockerfile, systemd unit or run script has a flag rename to make before the next bump — silent, and it fails at start-up rather than at build.

    llama.cpp·2026-08-15