model releases

Follow-up to last week's post about streaming MoE experts from disk in a browser tab. That ran on hand-written engines for two models. This week the same disk tier sits under llama.cpp's own WebGPU ba

Anthropic: Anthropic details Scion, a git replacement for its monorepo, built by seven engineers with Claude in five months, as daily merges rose from 2.5K in May to 30K+ — As our merges per day grew

hexagon: extend scalar DMA broadcasting support for binary ops (#30245) tests: cover weighted scalar broadcasts Assisted-by: OpenCode ggml : support extended scalar DMA binary ops Assisted-by: OpenCod

models : remove redundant ggml_cont calls now supported by all unary kernels (#30198) models: remove redundant ggml_cont calls now supported by CUDA unary kernels models: restore cont_2d after sigmoid

ggml-zdnn : introduce Q8_0 mat mul (#29836) ggml-zdnn: initial prototype of q8_0 matmul Signed-off-by: Aaron Teo aaron.teo1@ibm.com docs: update zdnn docs Signed-off-by: Aaron Teo aaron.teo1@ibm.com S

graph : avoid reallocations in the 'no cache' path (#30439) Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/attestations/55354972 macOS/iOS: macOS Apple Silicon (arm64)

cuda : add GGML_CUDA_GRAPH_DEBUG and read env vars once (#30327) ggml-cuda: gate CUDA graph debug logs behind GGML_CUDA_GRAPH_DEBUG ggml-cuda: cache env var lookups in static const instead of re-readi

vulkan: allow multiple devices to access gpu-visible host memory (#29741) vulkan: allow multiple devices to access gpu-visible host memory clean up erroneous newlines add Windows support handle nullpt

tests : consolidate context state tests into test-llama-context (#30023) tests : consolidate the context state tests into test-llama-context test-save-load-state, test-recurrent-state-rollback and tes

graph: skip the LM head for pooled embeddings (#30445) graph: skip the LM head for pooled embeddings A context that extracts pooled embeddings outputs one vector per sequence, but every model still ex

model: load YaRN beta metadata for laguna models (#30331) Assisted-by: Codex Co-authored-by: Adam Caldwell accaldwell@users.noreply.github.com Website: https://llama.app Attestations: https://github.c

llama : do not map foreign ranges into a backend buffer A buffer created from a host pointer pins everything that it wraps. When several contexts share the same file (lazy tensors, MoE tensors kept on

graph: pad the KQ bias of the no-cache attention (#30446) The no-cache attention pads K, V and the mask to 256, but T5 also adds a relative position bias of [n_tokens, n_tokens], so the add with KQ fa

hexagon: add configurable HTP DCVS/HMX power mode (#30423) hexagon: add configurable HTP DCVS/HMX power mode hexagon: address PR review feedback (drop CLI flag, align QAIRT power modes, comment/log cl

meta: recreate all subgraphs when the arena is reallocated (#30453) meta: recreate all subgraphs when the arena is reallocated chore: revert added test cases Website: https://llama.app Attestations: h

vulkan: Optimize pool_2d when the pooling window is large (#30119) When the window is larger than a subgroup, use a whole subgroup to compute each output. Website: https://llama.app Attestations: http

kleidiai: add SME2 Q4_K and Q6_K matmul support (#29219) Change-Id: Ib4aa4c9814c010f9cf084783cc80612e364f9a27 Change-Id: I406a6606c3fea51f8cd32331c1ed71466845d363 Change-Id: Ib4aa4c9814c010f9cf084783c

ggml : simplify backend search path error handling (#28367) Signed-off-by: Adrien Gallouët angt@huggingface.co Website: https://llama.app Attestations: https://github.com/ggml-org/llama.cpp/attestatio

Banger paper from Google DeepMind on co-evolving an agent's harness and its weights. (bookmark it) If you already log your harness changes and results, those logs can become training data for the mode

Dynamic vision, in action Basalt is an inference engine for 3.8 Flash Next, forked from Strata and heavily tuned for Blackwell. It supports vision, and for this to work you need to have the image enco

My project is optimising qwen flash next on Mac Mini m5 64gb through SSD streaming. I recently asked LocalLLama for guidance on how to optimise token output. Today a popular method to do this is a loo

I set up a dedicated workstation for local AI experimentation with various models and tools. But every single one of them wants all the available VRAM. To fix this, I built a request dispatching layer

I built a desktop app harness that keeps local meeting notes and can use them to autocomplete my typing as well as complement my dictation with additional context via reading my screen or using the pr

It's confirmed. OpenAI has been secretly replacing Astra with increasingly shit models since release. My evidence is overwhelming. Three Reddit comments, a benchmark that fluctuated 4%, and a guy who

@cjav_dev @aiDotEngineer @ricmac @latentspacepod our workshop writeup is live now! https://x.com/latentspacepod/status/2109713951604826153?s=20 At AI Engineer New York, Anthropic's @cjav_dev & @Harris

Hi Friends, I am a solo developer. Ever since I had Qwen 3.8 running on my RTX4090, I had an itch to use it on my phone for everything. That gave me the idea to build SteerClaw Mobile - a complete age

Sadly, all the tests were done using NanoGen, but maybe someone can try the idea on Krea 2 or Qwen-Image 2.1. Does our method work beyond ImageNet generation? Yes. We test it on NanoGen, a unified dif

Every few weeks, a new model steps up and claims best-in-class at something. The launch chart makes the case, the link makes the rounds, and your team is back to the question nobody has time for: shou

Quick note, not my character, original came from schmede_AI on youtube. But yes my question is do any local models currently have the ability to match and generate additional images of a character fro

Epic Games plans to use AI to generate code and art for an unannounced Disney game, which is now delayed from its original March release window, according to a report from Bloomberg. The project's dev

A few days ago I posted “Gausify Orbit - I am building a web application to transform images into Gaussian Splats”, and it was removed because the project wasn’t open source yet. It’s now released und

Last week we announced that Gemini Enterprise has evolved to make it even simpler to get work done, and introduced the Gemini agent, your new universal agent for work. It's always on and ready to answ

Gemini 4 Argon is available: — For developers: Build with the Gemini API paid tier and if you’re a Google AI Ultra subscriber, you can try it in @antigravity, @GoogleAIStudio, and @Android Studio. — F

Google LLC today brought its Gemini 4 Argon frontier artificial intelligence model to Gemini app users on its Pro and Ultra subscription plans, two weeks after the model went first to vetted cybersecu

After months sitting on the sidelines of the frontier AI race, Google has finally launched its long-awaited Gemini 4 Argon model, claiming it can compete with the best the industry has to offer. Googl

Robert Hart / The Verge: Google rolls out Gemini 4 Argon globally via the Gemini API and to Google AI Pro subscribers in the Gemini app and AI Mode, and in Antigravity for Ultra users — After months o

GPT-4 was the undisputed best model in the world for a full year and we all debated whether anyone could catch up or if OpenAI had some secret formula that nobody else knew about. Now there are are 19

Grok @Bot memory transfer NEW GROK BOT! AI MEMORY TRANSFER! Your old AI knows you. Your new one shouldn't have to start from zero. 🧠➡️ Meet Memory Transfer, a @Grok @Bot that brings over what Claude o

Grok @Bot rocks I've been using Grok @Bot as my daily driver for a few weeks now and: - it has replaced 100% of my search-like queries on chatgpt / claude - it has replaced 100% of my Hermes usage & a

So I wanted to try the horsey challenge myself on the latest engine i been testing which is a strata fork by sergqwer. https://github.com/sergqwer/strata-nvfp4. I seen a post earlier and seen i believ

A lot of projects have come up recently consisting of specific hardware config specialized forks of existing conferencing engines, but they all seem to be made using Claude or OpenAI's biggest frontie

https://preview.redd.it/m5e9anluo8vh1.png?width=1080&format=png&auto=webp&s=4c80687e6fe40036843a97ed8975ce46eaab0662 I am excited to share a project I have been working on during the past few years du

I spent a decent amount of time with Gemini 4 before launch. Overall, Gemini 4 is a big jump over previous Gemini models and it is the first non-OpenAI/Anthropic model to pass the Fable/Mythos/Astra b

If an agent saves an engineer 2 hours to implement a feature but collectively costs the team 3 hours in alignment and review, the ROI is negative. Isolated chats are suboptimal in team environments be

Hello everyone! My current setup has an RX 7800 XT 16GB GPU + 32GB RAM. I'm currently running Qwen3.8 27B at 4bit with max context 120k at around 40 t/s but the quality isn't that great, and I'd like

Laptops open across the room. Yesterday at AI Engineer New York, @cjav_dev + Harrison Stall of @AnthropicAI on Claude Managed Agents. Claude plans the work, then coordinates the agents. Workshop recap