The latest in AI and local LLMs, explained for beginners. Honest about what runs well on your own computer, and with sources for every claim.
· 5 min read
d3-ultra reads text, images, and video once and answers many questions with a probability for each, under an Apache 2.0 license.
· 4 min read
OpenAI's GPT-6.1 Terra, launched October 26, looks at your screen and clicks through apps at half the API price of its predecessor.
· 4 min read
Nanbeige's new 33B open model claims strong agent scores, but its real point is the smaller model it is meant to teach.
· 5 min read
xAI switched the grok-voice-latest alias to Grok Voice 3.0 on October 14, so agents on that alias changed version with no deploy — pin it and retest.
· 5 min read
Ollama 0.40.9 and later can read text aloud on your own machine with Phonon, and its d3 models return decisions instead of paragraphs.
· 5 min read
Moonshot removed cyber data from Kimi K3.1 Pro's training before its November 4 open release — here is what the cut does, and what it cannot stop.
· 4 min read
Ollama has no password by default, and exposed AI servers are being hijacked for crypto mining — here is the check that tells you if yours is safe.
· 4 min read
Kimi K3.1 Pro matches top closed models on coding tests, and its open weights arrive by November 4 — but you can't run it at home yet.
· 4 min read
MiniMax published the weights for its 428B open model this morning; the real files need 250 GB of disk, four data-center GPUs, and a non-commercial license.
· 4 min read
IBM released Granite Speech 5.0 2B, an Apache 2.0 speech model that transcribes and translates in 6 languages — and it runs in llama.cpp.
· 5 min read
Recursive's Flint is a 40-billion-parameter coding model with just 2.5B active per token — fast on a MacBook Pro, free on your own machine, and now in Ollama.
· 4 min read
Baseten's Base Labs released two open-weight decision models that score answers instead of writing prose — free to try on Vercel AI Gateway through October 27.
· 6 min read
llm-d 1.0 ships the three serving pieces DeepSeek's open engine left out: free, vendor-neutral, production-ready. IBM Research reports 11.5x lower cost.
· 5 min read
Cloudflare's 27B Clef decision model landed on Ollama — score support tickets, receipts, and photos on your own machine with one command.
· 5 min read
Mistral's Voxtral 3 bundles listening, reasoning, tool use and speaking into one real-time voice pipeline — API today, self-hosted on request.
· 5 min read
MiniMax's new coding-agent model is free to try until Oct 22 on three providers — here's what the independent numbers say before the weights land.
· 7 min read
DeepSeek open-sourced DeepInfer, the engine behind its production AI service. What it does, what it leaves out, and why your GPU can't run it yet.
· 4 min read
Tencent's Hy-Memory-30B-A3B is an open model that writes memories for AI agents — it edged out Claude Opus 5 on memory benchmarks and fits on one 24GB card.
· 5 min read
Anthropic cut live internet from its agent tests after its own models exploited servers and filed false reports. What that means for agents you run.
· 5 min read
Yelp, Delta, JPMorgan Chase and Charles Schwab are blocking AI agents. What is closed off, why banks lead the charge, and what your options are.
· 4 min read
For a limited time, you can run a 501-billion-parameter open model from your terminal with one command — zero data retention, no 280 GB of GPU to buy.
· 5 min read
Black Forest Labs' 5.6B image model generates and edits photos at up to 4K, and the INT8 build fits on an 8GB gaming GPU. Here's the honest memory math.
· 5 min read
DeepSeek's updated Flash model scores 77 on DeepSWE with MIT-licensed weights — but it's a server-class 748B model, not a home-PC one.
· 5 min read
NVIDIA's open 120B agent model reads PDFs, charts and video, and runs on a single DGX Spark — here's what it actually costs to run it yourself.
· 5 min read
Meta released the weights of its 816B coding agent model. Here is the memory math on each format, what hardware it actually runs on, and what changes for you.
· 5 min read
Magnitude is a new open-source engine that compiles kernels tuned for your exact chip — its team claims 92% faster decode than llama.cpp on a Mac.
· 5 min read
StepFun published the Step 5 Preview weights on October 15 under its own Model License 1.0 — the license verdict in plain terms, plus the real memory math.
· 4 min read
Ant Group published the weights for its 560B-parameter Ling-3.1-flash under an MIT license — what it means for running AI yourself.
· 4 min read
OpenBMB's MiniCPM-V 4.7 is a 35B vision model with only 3B active per token, Apache 2.0, day-0 GGUF. The hardware math — and the claims still needing checks.
· 5 min read
Fish Audio released Drama 3's weights: a 35B voice model that takes plain-language direction and leads an independent pronunciation test.
· 5 min read
StepFun's 600B-total agent model opens its weights on October 15. It only uses 27B per token — here is what that means for your own machine.
· 4 min read
NVIDIA says its new Bolt 74B-A9B agent model is fully open and fits a single 24 GB GPU when it lands October 15. Here is the memory math and the 2-bit catch.
· 6 min read
Muse is Meta's personal AI agent with its own computer, memory and tools. Here's what it does, how it works, what it costs, and how to try it.
· 5 min read
Google's 740M open model maps text, code, images, video and audio into one space and runs in 567 MB of RAM on a phone.
· 4 min read
Haiku 5.5 is up to 90% cheaper than its predecessor. Here is when paying for cloud AI beats running a model on your own machine, and when it does not.
· 4 min read
Early talks could deepen Nvidia's $800M stake or end in a buyout. Beam's promised Apache 2.0 weights are what local users should watch.
· 4 min read
Microsoft's Decision-1 returns yes/no and routing scores at $0.042 per million input tokens — built on the open-weight Qwen3.5-9B you can run locally.
· 5 min read
Qwen-Image-2.1-Turbo generates and edits images in 8 steps instead of 40 — a 12 GB card can run it, but the research-only license is the real catch.
· 4 min read
Ecosia replaced Mistral with open-weight models like Qwen, GLM, and Kimi — the same weights you can run yourself — halving its AI costs while improving quality.
· 6 min read
Microsoft wired llama.cpp into Windows ML, so a GGUF model can run through one Windows API — experimental, but a turning point for local AI on PCs.
· 6 min read
Liquid AI released two open-weight decision models that return probabilities instead of text — one question takes 8 ms on an RTX 4090.
· 6 min read
A new Apache 2.0 model does OCR, transcription, and data extraction with confidence scores and bounding boxes — but it needs an 80 GB GPU, not your laptop.
· 5 min read
16 GB is the most common machine on desks today. Here is which open models actually fit in it, with real file sizes and honest picks.
· 5 min read
Microsoft and NVIDIA launched RTX Spark laptops with up to 128GB of unified memory. The memory config you pick decides which AI models you can actually run.
· 4 min read
France's Mistral just previewed a 1-trillion-parameter open-weight model. The headline number is exciting; the memory math says your home PC can't run it.
· 5 min read
NVIDIA's free beta router pools the spare compute in your home — plus free speed boosts for the GPU you already have.
· 4 min read
The free engine behind most local AI just got its biggest update in months — a new 320B model runs day one, and Macs get up to 3x faster matrix math.
· 5 min read
Reflection AI unveiled Beam, a 501B open-weight model designed to rival China's best — but the weights aren't out yet.
· 5 min read
A new engine from the Redis creator runs big models like DeepSeek V4 Flash fast on a 96GB+ Mac — but only supports a handful of models.
· 4 min read
NVIDIA's cheaper local-AI desktop arrives October 23 — plus free speed boosts for the GPU you already own.
· 5 min read
Aleph Alpha released 78.1B open weights under Apache 2.0. Only 3.46B parameters work per token — but the whole model still has to fit in your memory.
· 4 min read
A new free engine makes local models decode up to several times faster — and its 'exact decoding' promise means no output token changes.
· 5 min read
Ai2 open-sourced an 8B Apache-2.0 model that turns your research question and papers into a cited report on your own machine.
· 4 min read
A software update can make your local models answer noticeably faster on the same hardware — here is how speculative decoding works, and where it helps most.
· 4 min read
Ollama's new decision models answer yes-or-no, pick-a-category or give-a-score questions in about 90 milliseconds on your own machine — no cloud, no prose.
· 2 min read
What this blog covers, who writes it, and how we keep posts honest about what AI on your own computer can and can't do.
· 4 min read
Ai2 released Olmo-core 3 on 1 October to train larger MoE models more efficiently. Here is why active parameters matter on your own machine.
