CompanyAnthropic8 recent entries9 Oct 2026Qwen 3.8 Flash Next-GSQ-RCO-IQ2_XS at ~21 tok/s on just an RTX 3060 12GB + 16GB DDR4 RAM(No gate pruning, 100% bit-exact)(Don't judge by the screenshot, the cache is cold. It hits 24+ tok/s with a warm cache!) About two months ago, I made a post here asking whether predicting which MoE experts would be used on the next →9 Oct 2026Mine Odyssey: Benchmarking Spatial Agentic Intelligence in the WildarXiv:2610.11328v1 Announce Type: new Abstract: Advances in foundation models are driving efforts to introduce agents to assist people in the physical world. Such agents require agentic spatial intell
CompanyOpenAI8 recent entries7 Oct 2026Semantic Capability Acquisition and Specialization During Vision-Language Model Fine-TuningarXiv:2610.07385v1 Announce Type: new Abstract: Fine-tuning vision-language models (VLMs) is typically evaluated at a single downstream checkpoint, obscuring whether a semantic capability was never ac→7 Oct 2026Rethinking Visual Provenance: Detection and Watermarking Across Direct Visual Generation and LLM-Driven Code RenderingarXiv:2610.08137v1 Announce Type: cross Abstract: AI systems create images and videos with image/video generation models or by writing code and graphics descriptions that are then rendered. These rout→8 Oct 2026I built a local server that speaks the Ollama API - sessions survive crashes, and your chat doesn't wait behind your agentsDisclosure first: I'm one of the maintainers, this is my project. If you've pointed Open WebUI or a script at a local server, you know the failure modes: the server restarts and every conversation is →9 Oct 2026rate the terminal setupik ik 'another harness D:'. so... i spent the last 6 months of my life building and ironing out every single detail and bug through almost every route with extensive testing across 3 machines, daily, →9 Oct 2026From Reactive Containment to Proactive Assurance: Lessons from OpenAI, Anthropic, and Google Agent Security IncidentsarXiv:2610.12463v1 Announce Type: cross Abstract: In 2026, cybersecurity evaluations involving OpenAI, Anthropic, and Google agents reached real systems outside their authorized test scope. The paths →13 Oct 2026open-source desktop agent app running its agents on local ollama models: two agents on qwen3.5:9b and qwen3.8 on a 16 gb card, real recording, including the limits i hiti'm the maintainer of dodaios (formerly personal jarvis), an open-source ai agent app for the desktop. apache 2.0, runs on windows, macos and linux, no account and no cloud of ours between you and you→13 Oct 2026One Agent Development Kit, five languagesWhen we talk about agents, that doesn't necessarily mean we're talking about Python. In today's modern ecosystem, an agent might run as a high-concurrency Go microservice on Cloud Run, or live inside →13 Oct 2026DigitalOcean Inference Router Enhancements: Performance Tuned Routing, No Rules to WriteEvery 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
CompanyGoogle8 recent entries6 Oct 2026Dynamic Harness Search: Building Multi-Agent Systems Per-Query via PredictionarXiv:2610.04137v1 Announce Type: new Abstract: Agent harnesses specify the roles, instructions, tools, and communication structure used to solve a task, and the right harness depends on the query. Be→7 Oct 2026Introducing Google Cloud’s U4 compute: Enabling ultra-low latency tradingIn capital markets, operational success is built on predictable performance and low latency, where even tiny variations in packet timing alter queue priority and trade execution. At the same time, bot→8 Oct 2026Welcome to Gemini at Work 2026: Introducing the Gemini agentEditor’s note: This article is adapted from Thomas Kurian’s keynote address at Gemini at Work 2026. It includes a number of announcements: Gemini, a single, universal agent for work that answers your →8 Oct 2026Empowering SMBs to do more with GeminiToday, we announced the new Gemini agent, your single, universal agent for work. For small and midsize businesses, with lean teams and margins, Gemini will be a critical tool for you to operate effici→9 Oct 2026Modernizing Unstructured Data Workflows: Alteryx Live Query meets Google Cloud BigQueryAlteryx One Live Query and Google Cloud BigQuery redefine how enterprises handle complex, unstructured data at scale by reducing reliance on disconnected tools, minimizing data movement, and enabling →9 Oct 2026From Reactive Containment to Proactive Assurance: Lessons from OpenAI, Anthropic, and Google Agent Security IncidentsarXiv:2610.12463v1 Announce Type: cross Abstract: In 2026, cybersecurity evaluations involving OpenAI, Anthropic, and Google agents reached real systems outside their authorized test scope. The paths →13 Oct 2026open-source desktop agent app running its agents on local ollama models: two agents on qwen3.5:9b and qwen3.8 on a 16 gb card, real recording, including the limits i hiti'm the maintainer of dodaios (formerly personal jarvis), an open-source ai agent app for the desktop. apache 2.0, runs on windows, macos and linux, no account and no cloud of ours between you and you→13 Oct 2026One Agent Development Kit, five languagesWhen we talk about agents, that doesn't necessarily mean we're talking about Python. In today's modern ecosystem, an agent might run as a high-concurrency Go microservice on Cloud Run, or live inside
CompanyMeta8 recent entries9 Oct 2026Rounding in Preconditioner Space: Redesigning 4-bit AdamW Optimizer-State QuantizationarXiv:2610.12444v1 Announce Type: new Abstract: Quantizing AdamW's optimizer states reduces persistent storage, but quantization errors propagate through the moment recurrences and perturb subsequent →9 Oct 2026My 'Anthropic/ChatGPT in a box' is... now open source and publicly hosted on Github.It's been a long time coming, and I should have done this in the first place. But LumaBrowser is now open source. Since I first released it a year ago, it's had several thousand installs and an absolu→11 Oct 2026Running Next Flash IQ3_XXS at ~70 tok/s with 100k context or 2 instances of Qwen 3.6 35B A3B IQ4 at ~145 tok/s with 256k all on $500 of ex mining BC-250 boardsThis will be my third update on the bc-250 cluster and for my first forray into local ai I have been having a blast. This has turned from a weekend experiment into something I think I will use in the →11 Oct 2026b11557: rpc : turn GGML_RPC_DEBUG into a verbosity level and add logs (#29544)rpc : turn GGML_RPC_DEBUG into a verbosity level and add logs GGML_RPC_DEBUG is now parsed as a number: 0/unset disables debug logs, 1-3 emit increasingly detailed output (events, per-command trace, t→12 Oct 2026v0.40.4What's Changed llama: update llama.cpp to b11521 for decision models and EmbeddingGemma 2 by @jmorganca in #18917 server: reuse verified manifests and loaded models in inference routes by @dhiltgen in→12 Oct 2026Taking Tokenization off the Host for Agentic LLM ServingarXiv:2610.13325v1 Announce Type: cross Abstract: Agentic workloads resend long contexts that barely change between calls. Prefix caching now skips most of their prefill, yet every request must still →12 Oct 2026strix-llama new branch: prompt processing +21–54% over Vulkan and up to 2.7× over stock ROCm kernel at 128k, same outputs qualityI've been tuning llama.cpp on my Strix Halo box (Radeon 8060S, 128 GB) and the numbers are good enough to share. It's a branch called rdna35-perf on top of halo-box/strix-llama.cpp, a llama.cpp fork a→12 Oct 2026Attention Sees, MLP Refuses: Tracing the Attention-to-MLP Safety Pathway in LLMsarXiv:2610.13078v1 Announce Type: new Abstract: Safety-aligned large language models (LLMs) remain vulnerable to jailbreaks and can over-refuse benign requests. Understanding these failures requires d
CompanyMistral8 recent entries14 Sept 2026Dead Weights, Live Signals: Feedforward Graphs of Frozen Language ModelsarXiv:2604.08335v2 Announce Type: replace Abstract: We present a feedforward graph architecture in which heterogeneous frozen large language models serve as computational nodes, communicating through →15 Sept 2026CrofAI 'cheapest inference provider in the world' gets exposed as an OpenRouter wrapper, routing requests to smaller, cheaper models at up to 20x markup. CrofAI responds to Wire Fraud allegations by denying everything, then backtracking, then 3 hours later wiping their entire online presenceDisclaimer: no AI was used whatsoever to write this post Cautionary tale about chasing cheap tokens. exposé: https://kendell.dev/blog/crofaifalse/ reaction by nahcrof, announcing the shutdown of the s→22 Sept 2026Written as a Record, Read as an Address: What a Forward Pass Leaves in an Operation's KV CachearXiv:2609.24635v1 Announce Type: new Abstract: When a language model reads an operation such as 'Swap the contents of Box F and Box B', its forward pass writes keys and values for those tokens into t→22 Sept 2026Machine-Interpretable Information: Compiling Documents into Searchable and Readable Protocol StatesarXiv:2609.23371v1 Announce Type: new Abstract: Long-context language models interface with external knowledge through raw natural language. In retrieval-augmented systems, this creates a persistent i→24 Sept 2026Can One Adapted Model Do It All? Fine-Tuning Strategy Selection for Customer Support LLMsarXiv:2609.27262v1 Announce Type: new Abstract: Production customer-support systems often require LLMs to support multiple skills, such as intent classification, question answering, summarization, or →5 Oct 2026WakeKV: Reactive, Reversible KV Residency for Heads That Change Their MindsarXiv:2610.02713v1 Announce Type: new Abstract: Most KV-cache compression methods classify attention heads once, either offline or during prefill, and keep this classification fixed throughout generat→8 Oct 2026Quad-State Safety Evaluation of Open-Weight Large Language Models on Non-Canonical InputsarXiv:2610.09033v1 Announce Type: new Abstract: Standard safety evaluations of large language models assess harmful requests written in canonical plain text, while models in real-world deployment rout→12 Oct 2026QLoRA on ExLlamaV3 Quants - Train @ 2-8bpwEXL3-QLoRA I've built this over the last few months and I think it's stable enough to share now. A fork of ExLlamaV3, this is a native ExLlamaV3 training implementation with plain PyTorch, without tra
CompanyxAI8 recent entries9 Sept 2026Wiki Lint Report — 2026-09-09Automated lint: 12 errors, 53990 warnings, 3 info→10 Sept 2026XAI-Refine: An Automated Explanation-Knowledge Loop for Brain-Age PredictionarXiv:2609.09388v1 Announce Type: new Abstract: Brain-age prediction models are commonly evaluated by predictive accuracy, yet accurate predictions alone do not establish that a model relies on reprod→13 Sept 2026Wiki Lint Report — 2026-09-13Automated lint: 12 errors, 54839 warnings, 3 info→15 Sept 2026Another Blueprint In The Wall: How to Ask Frontier AI Like a Kid?arXiv:2609.14803v1 Announce Type: new Abstract: This paper reports experiments across six frontier model types from OpenAI, Anthropic, xAI, and Google DeepMind. Ten independent sessions per model type→20 Sept 2026Wiki Lint Report — 2026-09-20Automated lint: 14 errors, 54840 warnings, 3 info→25 Sept 2026Watched my LLM router burn $11 on a 3-word prompt. Here's the fix.I run a personal always-on agent fleet — eight seats claiming work in parallel from a shared task board, routing across Anthropic, OpenAI, xAI, Google, and a local Ollama floor. This week the spend fo→27 Sept 2026Wiki Lint Report — 2026-09-27Automated lint: 17 errors, 59669 warnings, 3 info→4 Oct 2026Wiki Lint Report — 2026-10-04Automated lint: 19 errors, 64915 warnings, 3 info
CompanyDeepSeek8 recent entries7 Oct 2026SCOPE: Certified Theorem Proving with a Language Model as the Policy PlannerarXiv:2610.08319v1 Announce Type: new Abstract: In proof assistants such as Lean, a generated proof must pass machine compilation checks, so evaluation needs no human scoring. Direct generation fails →8 Oct 2026EngramEdit: Decoupled Knowledge Updates in LLMs through Conditional MemoryarXiv:2610.10533v1 Announce Type: new Abstract: Conditional memory architectures such as DeepSeek Engram use input n-grams to look up learned embeddings, expanding the capacity of large language model→8 Oct 2026$2800 rig with 8x Radeon Pro V620 (256 GB VRAM) + custom vLLM fork = Qwen3.8-Flash-Next at 60 to 100 t/s decode and 3000+ t/s prefillPost title is slightly misleading, I don't think you can get these for $350 each anymore but they're still pretty cheap all things considered. They're Radeon Pro V620's which are older RDNA2 enterpris→9 Oct 2026[Paper] EngramEdit: Decoupled Knowledge Updates in LLMs through Conditional MemoryConditional memory architectures such as DeepSeek Engram use input n-grams to look up learned embeddings, expanding the capacity of large language models (LLMs) with limited additional computation. Be→9 Oct 2026My 'Anthropic/ChatGPT in a box' is... now open source and publicly hosted on Github.It's been a long time coming, and I should have done this in the first place. But LumaBrowser is now open source. Since I first released it a year ago, it's had several thousand installs and an absolu→9 Oct 2026Mine Odyssey: Benchmarking Spatial Agentic Intelligence in the WildarXiv:2610.11328v1 Announce Type: new Abstract: Advances in foundation models are driving efforts to introduce agents to assist people in the physical world. Such agents require agentic spatial intell→12 Oct 2026Taking Tokenization off the Host for Agentic LLM ServingarXiv:2610.13325v1 Announce Type: cross Abstract: Agentic workloads resend long contexts that barely change between calls. Prefix caching now skips most of their prefill, yet every request must still →13 Oct 2026DigitalOcean Inference Router Enhancements: Performance Tuned Routing, No Rules to WriteEvery 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
CompanyNVIDIA8 recent entries8 Oct 2026Building Reliable Data Analytics Agents: Lessons from the KDD CupThe NVIDIA KGMON team secured second place in the KDD Cup 2026 Data Agents competition by building a constrained harness around a small fixed LLM, unifying heterogeneous data into an SQLite database w→9 Oct 2026Super interesting NVIDIA paper on choosing base models for coding agents. It's actually a clever way to rank base checkpoints by how well ea…Super interesting NVIDIA paper on choosing base models for coding agents. It's actually a clever way to rank base checkpoints by how well each one is likely to do as a coding agent after post-training→9 Oct 2026DivMoE: Fine-Grained MoE Upcycling via Cross-Domain Expert CompositionarXiv:2610.11317v1 Announce Type: new Abstract: Mixture-of-Experts (MoE) architectures have become essential for scaling large language models, with recent work demonstrating the benefits of fine-grai→11 Oct 2026vllm-ascend updates (GLM5.3-flash) on dual 310p Ascend cardsAlright, I learend a few valuable lessons since I posted about a week ago, and I've made more strides with the DDR4x Ascend 96GB 310P cards I purchased, so here we go, meet the new friendly and less v→12 Oct 2026Taking Tokenization off the Host for Agentic LLM ServingarXiv:2610.13325v1 Announce Type: cross Abstract: Agentic workloads resend long contexts that barely change between calls. Prefix caching now skips most of their prefill, yet every request must still →12 Oct 2026RAMP: Recurrent Adaptive Mixed Precision for Efficient Prefill in Recurrent-Depth Language ModelsarXiv:2610.13041v1 Announce Type: new Abstract: Recurrent-depth language models improve inference-time reasoning by repeatedly applying a shared network to refine internal representations, but the add→13 Oct 2026open-source desktop agent app running its agents on local ollama models: two agents on qwen3.5:9b and qwen3.8 on a 16 gb card, real recording, including the limits i hiti'm the maintainer of dodaios (formerly personal jarvis), an open-source ai agent app for the desktop. apache 2.0, runs on windows, macos and linux, no account and no cloud of ours between you and you→13 Oct 2026LiquidAI/LFM2.5-24B-A2BGguf's: https://huggingface.co/LiquidAI/LFM2.5-24B-A2B-GGUF Free Demo: https://playground.liquid.ai 'LFM2.5-24B-A2B is a 24B-parameter reasoning model with only 2B active parameters, built for fast, p