News

New models, hardware, research, and techniques for running AI locally. We are selective: a quieter feed of things that change what is possible, practical, or legal is more useful than a complete one.

27 October 2026 Model release

MiniMax releases M3.1 Flash, a 428B open-weight MoE under a conditional commercial licence

MiniMax has published the weights for M3.1 Flash, a multimodal mixture-of-experts model of roughly 428 billion total parameters (about 23 billion active per token) with a context window up to one million tokens, downloadable now in 8-bit. It is open weights rather than open source: the MiniMax Community Licence allows commercial use, but asks for attribution and a one-time notice, and requires separate written authorisation above 20 million US dollars in annual revenue.
27 October 2026 Ecosystem

Ollama can now generate speech on macOS, and adds the d3 decision models

Ollama's 0.40.11 release adds local text-to-speech on macOS: ollama run can read a chat model's reply aloud as the audio streams, and apps can call an OpenAI-compatible speech endpoint. It also adds the d3 family of small decision models in 2B, 4B, 9B and 27B sizes, which classify, answer yes or no, or score options. The speech model and the d3 models arrive with provenance and licence details Ollama has not yet documented, so treat those as unconfirmed.
23 October 2026 Policy

China is drafting export controls for frontier AI models, not the open weights you already run

China's commerce ministry has published a draft that would require an export licence for the technology behind the very largest AI models, and ban it outright for models trained on sensitive state data. The rules are open for comment until 15 November and are not yet law, though a pre-export registration step took effect immediately. For anyone running Chinese open-weight models locally, the practical news is reassuring: weights already released as open source, and access over an API or on a device, are explicitly outside the scope.
23 October 2026 Model release

A games studio built Raon-1 mini, an open-weight MoE that targets a single RTX 5090

KRAFTON, the Korean studio behind PUBG, has released a preview of Raon-1 mini, a 30-billion-parameter mixture-of-experts model it trained from scratch and put out under Apache 2.0 behind a sign-in gate. A 4-bit build is designed to run on one RTX 5090, and it ships with a draft model for speculative decoding. The self-reported benchmarks are strong on maths and reasoning but trail peers on agentic and tool-use work, and it is a preview, so the numbers are the publisher's own.
21 October 2026 Model release

Reflection AI releases Beam, a 501-billion-parameter open-weight reasoning model under Apache 2.0

Reflection AI has released the weights for Beam, a 501-billion-parameter mixture-of-experts reasoning model, under a genuinely permissive Apache 2.0 licence. It is a notable US-origin frontier open model, though at this size it is firmly server and high-end-unified-memory territory, not a consumer-card one.
21 October 2026 Policy

The US has ordered DeepSeek out of federal agencies, not off your machine

A US Department of Homeland Security order bars DeepSeek from federal civilian agencies and their procurement. Despite the "US bans DeepSeek" headlines, it does nothing to a person or company downloading and running DeepSeek's open weights on their own hardware, which is a distinction worth being precise about.
12 October 2026 Model release

JetBrains releases Mellum 2.1 Thinking, a small Apache-licensed coding model built to run locally

JetBrains has released Mellum 2.1 Thinking, a 12-billion-parameter mixture-of-experts coding model with only 2.5 billion active parameters, an Apache 2.0 licence, and a 131k context. It is small enough to run on modest hardware, and its self-reported coding scores are strong, though one benchmark on its own card tells a more mixed story.
6 October 2026 Model release

Mistral unveils Large 4, a 1.05-trillion-parameter open-weight model, with the weights still to come

Mistral has announced Large 4, nicknamed "Le Chonk": a 1.05-trillion-parameter multimodal mixture-of-experts it says rivals the strongest open models. The API preview is live now, but the open weights are not due until the end of October, and Mistral has not yet stated the licence they will carry.
6 October 2026 Ecosystem

Ollama makes MLX the default runtime on Apple Silicon

Ollama's v0.40.0 release runs supported model architectures on Apple's MLX framework by default on Apple Silicon, rather than through its usual llama.cpp path, and broadens MLX support to more recent models. On a Mac, that can mean better use of unified memory with no manual configuration.
28 September 2026 Model release

Tencent's Hunyuan 4 flagship ships as an Apache 2.0 open model, with no territory clause

Tencent has open-weighted Hy4-preview, its Hunyuan 4 flagship: a 770-billion-parameter text mixture-of-experts with 49 billion active per token and a 1-million-token context. Read at source, the licence is standard Apache 2.0 with no geographic restriction, which makes it usable in the UK, unlike the Hunyuan media models we flagged earlier. It is server-class, but genuinely and globally open.
Featured
10 September 2026 Model release

DeepSeek releases V4.1 Flash, an MIT-licensed multimodal model with an unusual architecture

DeepSeek has published V4.1 Flash under the MIT licence: a 552-billion-parameter multimodal mixture-of-experts that reads images and text and activates only 8 to 16 billion parameters per token. It uses an encoder-decoder design rather than the decoder-only shape almost every other current model uses, and it carries a 1-million-token context. It is server-class, but genuinely open.
Featured
10 September 2026 Policy

Several open video models exclude UK users by licence, outputs included

MiniMax H3, a widely shared open-weight video model, is licensed for use everywhere except the EU, the UK, Korea and the US, and the restriction covers the model's outputs as well as its weights. It is not alone. For a UK reader, this is a case where the licence, not the hardware, is what stops you, and it is invisible to a licence badge.
3 September 2026 Ecosystem

NVIDIA upstreams llama.cpp optimisations it says make local inference up to 1.9x faster

At IFA 2026 NVIDIA announced local-AI updates, and the quietly important part is software: kernel optimisations it has contributed to the open-source llama.cpp and vLLM projects, which it says lift throughput by up to 1.9x on a GeForce RTX 5090. Because they are upstream, existing NVIDIA GPU owners get the gain by updating, with no new hardware. A new RTX Spark desktop with 128GB of unified memory was also shown.
2 September 2026 Model release

GLM-5.3 open weights land, under a bespoke licence rather than MIT

Z.ai has published the open weights of GLM-5.3, roughly two weeks after its hosted launch and following a security review. The full model is a server-class mixture-of-experts of around 744 billion parameters under a bespoke GLM-5.3 licence, not the MIT terms its predecessor carried. A smaller, natively multimodal GLM-5.3-Flash also shipped, and that one is MIT.
Featured
25 August 2026 Hardware

Apple's M5 Mac Studio and Mac mini refresh puts 512GB of unified memory on the desk

Apple has announced M5-generation Mac Studio and Mac mini machines and is positioning them squarely at local AI. The headline for people running large models is the top Mac Studio: an M5 Ultra with up to 512GB of unified memory at a claimed 1.2TB/s, enough to hold very large open-weight models in memory on a single desktop. Pre-orders opened on 25 August, with machines shipping 22 September.
Featured
18 August 2026 Ecosystem

Local generative media grows up: open video and music you can actually run

Generating video and music locally used to mean data-centre hardware or a non-commercial licence. That is changing: there are now open video models that fit a single 24GB card and permissively-licensed music models that run in 8GB. A short survey of where local generative media stands, and the licence traps to watch.
15 August 2026 Model release

Zhipu launches GLM-5.3 with strong coding claims, but the weights are not out yet

Zhipu has released GLM-5.3 through its hosted coding service, claiming the strongest open-weight coding model available. Two caveats matter for a local audience: the weights are not downloadable yet, said to follow in roughly two weeks after a security review, and every benchmark figure so far is the company's own. We will cover it properly, and add it to the catalogue, when the weights are actually released.
14 August 2026 Model release

Qwen3.8 arrives in the open: a runnable 27B, and a text-only 2.4T flagship

The Qwen3.8 open weights our earlier report treated as pending have now landed, and they split cleanly in two. Qwen3.8-27B is a permissively licensed, multimodal model that runs on a single 24GB card. The 2.4-trillion-parameter flagship is a text-only, cluster-scale release under a custom licence. For most people running local AI, the 27B is the story.
Featured
12 August 2026 Model release

Liquid AI releases LFM2.5-VL-3B, an open-weight vision model built to run on-device

Liquid AI has released LFM2.5-VL-3B, a 3.1-billion-parameter open-weight vision-language model designed to run on phones, laptops, and single GPUs rather than in a data centre. It ships with day-one GGUF, MLX, and ONNX exports and support across the common local runtimes, and adds screen understanding, object grounding, and function calling. The licence is open but carries a revenue condition on commercial use.
11 August 2026 Model release

NVIDIA releases Nemotron 3.5 Lightning, a fully open 30B mixture-of-experts

NVIDIA has released Nemotron 3.5 Lightning, an open 30-billion-parameter mixture-of-experts with about 3.6 billion active per token, a hybrid Mamba-Transformer design, and a one-million-token context. It is unusually open, shipping under a permissive licence with the training data and recipes alongside the weights, and it is aimed at fast, high-volume agentic work. Runtime support for its hybrid architecture is still maturing in local engines.
10 August 2026 Model release

Meta releases Muse Glimmer, a 30B open model under Apache 2.0 built to run on one GPU

Meta has released Muse Glimmer-30B, a multimodal model distilled from its larger closed Muse Spark and aimed at agentic use on a single consumer GPU. The headline for local AI is the licence: it ships under Apache 2.0, dropping the conditions that Meta's Llama models carry.
4 August 2026 Ecosystem

Ollama's August update speeds up Apple Silicon inference with speculative decoding

Ollama v0.32.6 has the MLX engine use a model's own multi-token-prediction head to drive speculative decoding automatically, which speeds up Qwen on Apple GPUs. It also aligns streaming with OpenAI's format and widens GPU support. An incremental release, with a couple of genuinely useful improvements.
3 August 2026 Model release

Alibaba releases Qwen3.8-Max as open weights, with a runnable 27B to follow

Alibaba has announced Qwen3.8-Max, a 2.4-trillion-parameter mixture-of-experts model, with open weights said to be arriving on Hugging Face, alongside a much smaller Qwen3.8-27B for ordinary hardware. As of writing the weights are not yet published and no licence has been named, so treat both as announced rather than available.
27 July 2026 Model release

Moonshot releases Kimi K3, a 2.8-trillion-parameter open model for server-class self-hosting

Moonshot AI has released the open weights for Kimi K3, a 2.8-trillion-parameter mixture-of-experts model that tops several open-weight leaderboards. It is a clear case of self-hosting at server scale: running it needs a cluster, not a workstation, and its licence terms were not clearly published at launch.
6 July 2026 Model release

Mistral releases the Mistral 3 family under Apache 2.0

Mistral has released the Mistral 3 family, spanning small dense models and a large mixture-of-experts, all under the Apache 2.0 licence. The licensing is as notable as the models: it removes the commercial conditions of Mistral's earlier releases.
2 July 2026 Technique

Mistral open-sources Leanstral, a model built for formal verification

Mistral has released Leanstral 1.5, an Apache 2.0 model specialised for Lean 4, the proof assistant used in formal software verification. It is a reminder that local models are not only general chatbots, but increasingly tools for specific domains.
29 June 2026 Policy

EU finalises Digital Omnibus, delaying the AI Act's high-risk rules to 2027

The Council of the EU has given final approval to the Digital Omnibus, a package that simplifies and defers parts of the AI Act. The main high-risk obligations move from August 2026 to December 2027, while the prohibitions and model rules stay live.
Featured
18 May 2026 Hardware

Apple trims high-memory Mac Studio options amid a DRAM squeeze

A tightening memory market has led Apple to narrow the memory configurations on its top Mac Studio, removing some of the very high-capacity options that made it a favourite for running large models. If you were counting on a specific config, check current availability.
24 April 2026 Model release

DeepSeek releases V4 as open weights under MIT

DeepSeek has released its V4 generation, reportedly as two models, V4-Pro and V4-Flash, under the permissive MIT licence and as open weights. The company frames it as an architectural step rather than a scale-up. Treat the performance claims as claims until independent results appear.
16 April 2026 Model release

Qwen3.6 arrives with new mixture-of-experts checkpoints

Alibaba has followed Qwen3.5 with Qwen3.6, reported to include new open-weight checkpoints, among them a mixture-of-experts model. The rapid cadence continues to be as notable as any single release in the line.
10 March 2026 Hardware

Strix Halo mini PCs put 128GB of unified memory within reach for local AI

Mini PCs built on AMD's Ryzen AI Max+ 395 pair a large integrated GPU with up to 128GB of unified memory, most of it usable as VRAM. They run models that no single consumer discrete card can hold, at low power, though bandwidth caps their speed.
16 February 2026 Model release

Alibaba releases Qwen3.5, extending its open-weight line

Alibaba's Qwen team has released the Qwen3.5 series, continuing the family's fast release cadence and its Apache 2.0 licensing. As always, treat the accompanying benchmark figures as the developer's own claims until independent results appear.
15 January 2026 Model release

Black Forest Labs releases FLUX.2 [klein] compact image models

Black Forest Labs has released FLUX.2 [klein], a family of compact image models, reported at 4B and 9B sizes, built for fast, near-real-time generation on consumer GPUs. It is the openly licensed member of the new FLUX.2 line.
15 October 2025 Hardware

NVIDIA's DGX Spark launches, a 128GB desktop AI machine

NVIDIA's DGX Spark, a compact desktop built on the GB10 Grace Blackwell chip with 128GB of unified memory, has gone on sale through NVIDIA and partners. It targets developers who want NVIDIA's software stack on a local machine, with modest bandwidth as the main trade-off.
6 December 2024 Model release

Meta releases Llama 3.3 70B, matching much larger models at lower cost

Meta has released Llama 3.3 70B Instruct, an open-weight model that the company says approaches the quality of its 405B model while needing far less hardware to run. It ships with a 128k context window under the Llama 3.3 Community License.

Showing all articles. As the archive grows this listing will paginate.