CompanyAnthropic8 recent entries19 Sept 2026Steer LLMs and Agents at the Token Level: An interactive tool for token visualization & control, model inspection and data annotation.onPanda is designed for geeks, power users, curious minds, and engineers. Its UI is built for deep exploration and efficient data annotation. - The core loop is simple: hover over a token → click an a→25 Sept 2026The agent-first cloud: why we built Managed AgentsCompute (EC2) opened the front door to the first generation cloud. Agents need something different: compute, inference, and data working as one. We built the cloud that delivers all three as one simpl
CompanyOpenAI8 recent entries16 Sept 2026Write up on building and improving reliability of agentic harnesses + a big dataset from my work.I've been working on something for this community (and other self-hosted communities) for almost 5 months now. Personally, I would grade it as probably the closest open source equivalent of Manus and →23 Sept 2026v0.5.0Overview This release focuses on backend performance and correctness, broader model coverage, and more robust server/router operation. It adds HRM-Text (DFM Mimir 1B) support, MiMo-V2.6 and HunyuanOCR→28 Sept 2026Muslim: A Deployed Arabic Voice AI Platform for Grounded Islamic KnowledgearXiv:2609.31511v1 Announce Type: new Abstract: We present Muslim, a production Arabic voice AI platform serving grounded, sourced Islamic knowledge to real users. Beyond a real-time voice pipeline (N→28 Sept 2026DeepEdu-v1: Efficient and Scalable Agentic LLMs for Vietnamese EducationarXiv:2609.31568v1 Announce Type: new Abstract: AI tutoring could markedly improve learning outcomes for students in developing regions such as Vietnam, yet the two obvious paths both fall short. Clou→29 Sept 2026Gentle reminder that Hugging Face was declined help from Anthropic & OpenAI models when they were being hacked by OpenAI models Self-hosted …Gentle reminder that Hugging Face was declined help from Anthropic & OpenAI models when they were being hacked by OpenAI models Self-hosted GLM 5.2 helped them protect themselves when the closed labs →1 Oct 2026Computer use powered by local/cloud models for regulated industries?The local models seem powerful enough to be capable of running basic local computer use. This is a computer use agent/harness built via claude code, and powered by qwen3.8 flash next nvfp4. Drawing a →3 Oct 2026Anyworld, a self-hosted multiplayer text RPG where a local LLM is the Dungeon MasterHey everyone, I’ve been working on a game called Anyworld. It’s a browser-based multiplayer (single player also supported) text adventure inspired by the early days of AI Dungeon, especially its brows→7 Oct 2026Running an open-source agent platform with no third party in the chain: what each piece needs (Ollama/vLLM, local embeddings, local OCR)Disclosure first: I work at Vstorm and I build AgenticOS, an open-source (Apache-2.0), self-hosted platform for building AI agents in a browser. This post was drafted with an LLM's help and edited by
CompanyGoogle8 recent entries10 Aug 2026Meta returns to open weights: Muse Glimmer, its first open-weights release since Llama 4, scores 35 on the Artificial Analysis Intelligence …Meta returns to open weights: Muse Glimmer, its first open-weights release since Llama 4, scores 35 on the Artificial Analysis Intelligence Index. It is a 30B-parameter model, and the first from Meta →30 Aug 2026We used HFlow to evaluate the latest open weights VLMs for processing egocentric dataWe used HFlow to evaluate the latest open weights VLMs for processing egocentric data. This was based on Build AI's Egocentric-10k evaluation, which used Gemini 2.5 Flash to measure hand visibility an→2 Sept 2026Value analysis of the old Ollama Pro sub model vs the new per-token model (from 50 days of real usage)Why I have this data: one Pro account kept hitting limits, so I run three Pro subs through 9router, a self-hosted router that sits between my coding tools and ollama cloud. It spreads requests across →16 Sept 2026Write up on building and improving reliability of agentic harnesses + a big dataset from my work.I've been working on something for this community (and other self-hosted communities) for almost 5 months now. Personally, I would grade it as probably the closest open source equivalent of Manus and →17 Sept 2026How a solo founder runs a five-continent tender platform on AlloyDB and MCPEditor's note: Lucius AI, a tender-intelligence startup covering markets across five continents, runs its entire data platform on AlloyDB for PostgreSQL with a single operator. By migrating semantic s→21 Sept 2026Got jev-like api running natively on ninfer / qwen3.8 27b and results are quite decentHi all! After all the jev hype I wanted to try by myself and got qwen3.8 27b to mix-and-match open implementations on top of ninfer, and to be honest I'm quite impressed with the results, latency is n→23 Sept 2026Which small model works best with Ollama on a Raspberry Pi 5?I've been comparing and working on Gemma 4, Phi-4-Mini, and Qwen3-4B for local LLM use on an 8 GB Raspberry Pi 5. The results were indeed close, but each model has a different strength: Gemma 4 E2B: a→28 Sept 2026Why your startup needs open models alongside frontier APIsEvery week, I talk with founders who are building at an unbelievable pace. Teams are moving from inception to product-market fit faster than ever, with foundation models wired deeply into their core p
CompanyMeta8 recent entries10 Aug 2026Meta returns to open weights: Muse Glimmer, its first open-weights release since Llama 4, scores 35 on the Artificial Analysis Intelligence …Meta returns to open weights: Muse Glimmer, its first open-weights release since Llama 4, scores 35 on the Artificial Analysis Intelligence Index. It is a 30B-parameter model, and the first from Meta →11 Aug 2026I built a weird, low-power llama.cpp server using an Intel N100 + RTX 5060TiEverything started with the sudden death of my old ASRock J1900. While looking for the perfect ITX replacement, I stumbled upon the Chinese CW-NAS-ADLN-K motherboard, which looked perfect on paper: In→12 Aug 2026How OneAdvanced deployed over 50 AI agents on UK-sovereign AWSLearn how OneAdvanced, a UK enterprise software provider, built a UK-sovereign AI platform by self-hosting Llama 4 Maverick and Llama Guard 4 on Amazon SageMaker AI, with a RAG pipeline on pgvector an→19 Aug 2026New open-source agent harness just landed! I got early access to TrueForge by TrueFoundry and have been running it locally for the past few …New open-source agent harness just landed! I got early access to TrueForge by TrueFoundry and have been running it locally for the past few days. The harness layer deserves as much attention as the mo→14 Sept 2026v0.4.1Overview llama.cpp 0.4.1 adds Maple 20B-A1B, Tencent Hy 4, and Spark2.5 support, improves JSON schema handling, chat parsing, logging, and server child-process management, and updates ggml to v0.24.0.→16 Sept 2026b10993ci : add self-hosted webgpu to hf-jobs (#28712) add self-hosted vulkan and webgpu to hf-jobs try t4-medium cont : adjust cpu backend threads try t4-small again restore cm jobs Co-authored-by: Georgi G→23 Sept 2026v0.5.0Overview This release focuses on backend performance and correctness, broader model coverage, and more robust server/router operation. It adds HRM-Text (DFM Mimir 1B) support, MiMo-V2.6 and HunyuanOCR→25 Sept 2026Combating Instruction Conflict via Energy-Driven Latent Conflict DetectionarXiv:2609.08646v2 Announce Type: replace Abstract: Large Language Models (LLMs) are increasingly deployed with hierarchical instructions, yet they remain vulnerable to conflicts in which user directi
CompanyMistral6 recent entries23 Jun 2026Available today: the API, Document AI in Mistral AI Studio, Amazon SageMaker, Microsoft Foundry, coming soon Snowflake Parse Document, or se…Available today: the API, Document AI in Mistral AI Studio, Amazon SageMaker, Microsoft Foundry, coming soon Snowflake Parse Document, or self-hosted on a single container, so your documents never lea→2 Jul 2026GRACE-RAG: Governed Retrieval Architecture for Canonical Evidence Synthesis, Enabling Lightweight Deployment in Closed-Domain Institutional SettingsarXiv:2607.00013v1 Announce Type: cross Abstract: Retrieval-Augmented Generation (RAG) systems are widely used in institutional question answering settings where responses must be grounded in authorit→31 Jul 2026What’s new in AI infrastructure and orchestration this monthAt Google, AI is a soup-to-nuts endeavor. Obviously, we make leading AI models like Gemini and Nano Banana. We incorporate AI into the tools you use every day (think Gmail, BigQuery, AlloyDB, Google C→3 Aug 2026I gave five different local LLMs a town. They invented Facebook and a duck-based credit bureau. (MIT, self-hosted, you don't play it — you watch it)Each villager in Pepperton is a different model — a mistral, a qwen3, a qwen2.5, a phi4-mini, a llama3.2 — because model families have genuinely different temperaments, and the friction between them i→19 Aug 2026GLM-5.3 is out on AA, and I'm fed up with their Intelligence/cost plotI think AA's intelligence/cost plot is seriously misleading, so I decided to make my own. Their plot is in the second image. All points are at max thinking. All intelligence index scores are from AA. →25 Sept 2026Combating Instruction Conflict via Energy-Driven Latent Conflict DetectionarXiv:2609.08646v2 Announce Type: replace Abstract: Large Language Models (LLMs) are increasingly deployed with hierarchical instructions, yet they remain vulnerable to conflicts in which user directi
CompanyxAI1 recent entries18 Aug 2026Local Qwen 3.8 27B vs GPT‑5.6 Terra vs Grok 4.6I gave three AI models the same brief: build a premium Three.js fragrance launch site from the same Git baseline, independently and with no collaboration. Three very different results. Here’s the full
CompanyDeepSeek8 recent entries19 Aug 2026GLM-5.3 is out on AA, and I'm fed up with their Intelligence/cost plotI think AA's intelligence/cost plot is seriously misleading, so I decided to make my own. Their plot is in the second image. All points are at max thinking. All intelligence index scores are from AA. →28 Aug 2026Frona v2026.8.0 – self-hosted personal AI assistant with ontology memoryHey, An LLM writing into durable memory is an unreliable author. It invents, mislabels, restates, and contradicts. Once one of those mistakes is saved, the agent can carry it into every conversation t→2 Sept 2026Value analysis of the old Ollama Pro sub model vs the new per-token model (from 50 days of real usage)Why I have this data: one Pro account kept hitting limits, so I run three Pro subs through 9router, a self-hosted router that sits between my coding tools and ollama cloud. It spreads requests across →3 Sept 2026We built an open-source, model-neutral agent harness and compared it with claude managed agents - for the same model, got same accuracy, upto 75% lower costWe have been working on an open-source, model-neutral agent harness for general purpose agents called TrueForge, and wanted to understand how much the harness itself actually matters. So we ran 14 tas→14 Sept 2026v0.4.1Overview llama.cpp 0.4.1 adds Maple 20B-A1B, Tencent Hy 4, and Spark2.5 support, improves JSON schema handling, chat parsing, logging, and server child-process management, and updates ggml to v0.24.0.→21 Sept 2026Got jev-like api running natively on ninfer / qwen3.8 27b and results are quite decentHi all! After all the jev hype I wanted to try by myself and got qwen3.8 27b to mix-and-match open implementations on top of ninfer, and to be honest I'm quite impressed with the results, latency is n→23 Sept 2026v0.5.0Overview This release focuses on backend performance and correctness, broader model coverage, and more robust server/router operation. It adds HRM-Text (DFM Mimir 1B) support, MiMo-V2.6 and HunyuanOCR→12 Oct 2026Locate, Judge, Read: Role-Specialized Tool Agents over a Knowledge-Point Index for Open-Domain Multimodal Document QAarXiv:2610.13128v1 Announce Type: new Abstract: Question answering over large heterogeneous document collections requires first finding the right passages and then reading them, yet current retrieval-
CompanyNVIDIA8 recent entries11 Aug 2026I built a weird, low-power llama.cpp server using an Intel N100 + RTX 5060TiEverything started with the sudden death of my old ASRock J1900. While looking for the perfect ITX replacement, I stumbled upon the Chinese CW-NAS-ADLN-K motherboard, which looked perfect on paper: In→18 Aug 2026The Commercial Tax: Rent-vs-Own Blind Spots in Multi-Hop Retrieval BenchmarksarXiv:2608.16096v1 Announce Type: cross Abstract: Enterprises connect language models to their own data through retrieval. The benchmarks that rank multi-hop retrieval systems leave out two facts a bu→27 Aug 2026Deepgram deepens Amazon SageMaker AI observability with Enhanced MetricsSelf-hosted speech AI carries an observability trade-off: the numbers that drive capacity planning and cost management stay locked inside the vendor container. Deepgram closes that gap on Amazon SageM→9 Sept 2026A Measurement Study of LLM Inference Trade-offs Across Edge Continuum HardwarearXiv:2609.08307v1 Announce Type: cross Abstract: Large language models (LLMs) are increasingly used as backends for intelligent web services, but serving them across the edge continuum requires balan→14 Sept 2026v0.4.1Overview llama.cpp 0.4.1 adds Maple 20B-A1B, Tencent Hy 4, and Spark2.5 support, improves JSON schema handling, chat parsing, logging, and server child-process management, and updates ggml to v0.24.0.→23 Sept 2026v0.5.0Overview This release focuses on backend performance and correctness, broader model coverage, and more robust server/router operation. It adds HRM-Text (DFM Mimir 1B) support, MiMo-V2.6 and HunyuanOCR→28 Sept 2026Why your startup needs open models alongside frontier APIsEvery week, I talk with founders who are building at an unbelievable pace. Teams are moving from inception to product-market fit faster than ever, with foundation models wired deeply into their core p→6 Oct 2026CoreWeave targets GPU utilization in continuous AI post-trainingGPU utilization during post-training depends partly on how efficiently infrastructure moves data and loads updated models. As enterprises continually refine AI agents, reducing delays between training