---
title: "AlphaSignal Llms News"
url: "https://alphasignal.ai/topic/llms"
description: "Latest AI and machine learning news about Llms on AlphaSignal."
---

# AlphaSignal Llms News

Latest AI and machine learning news about Llms on AlphaSignal.

## Articles

- [Nagoya University's Ruri v3 Beats OpenAI Embeddings on Japanese Retrieval](https://alphasignal.ai/news/nagoya-university-s-ruri-v3-beats-openai-embeddings-on-japanese-retrieval.md) (2026-10-15)
- [MBZUAI Ships Jais 3, Six Open Arabic Models Beating Qwen and Claude](https://alphasignal.ai/news/mbzuai-ships-jais-3-six-open-arabic-models-beating-qwen-and-claude.md) (2026-10-15)
- [Baidu's ProxyAttn Cuts Long-Context AI Processing 10x Without Retraining](https://alphasignal.ai/news/baidu-s-proxyattn-cuts-long-context-ai-processing-10x-without-retraining.md) (2026-10-15)
- [TypeSafe AI's Jev 1.14 Brings Vision and 256K Context to Typed Decisions](https://alphasignal.ai/news/typesafe-ai-s-jev-1-14-brings-vision-and-256k-context-to-typed-decisions.md) (2026-10-15)
- [Google DeepMind Shrinks Gemma 4 E4B to 5GB Without Losing Reasoning Power](https://alphasignal.ai/news/google-deepmind-shrinks-gemma-4-e4b-to-5gb-without-losing-reasoning-power.md) (2026-10-14)
- [Japan's NII Releases LLM-jp-4 to Beat GPT-4o on Japanese Benchmarks](https://alphasignal.ai/news/japan-s-nii-releases-llm-jp-4-to-beat-gpt-4o-on-japanese-benchmarks.md) (2026-10-14)
- [Cursor Ships Gemini 4 Argon With a 1M-Token Output Limit](https://alphasignal.ai/news/cursor-ships-gemini-4-argon-with-a-1m-token-output-limit.md) (2026-10-13)
- [Liquid AI's LFM2.5-24B-A2B Runs 184 Tokens per Second on a Laptop](https://alphasignal.ai/news/liquid-ai-s-lfm2-5-24b-a2b-runs-184-tokens-per-second-on-a-laptop.md) (2026-10-13)
- [Artificial Analysis' Decision Index Shows Jev Beats Claude Haiku 5.5 at 11x Speed](https://alphasignal.ai/news/artificial-analysis-decision-index-shows-jev-beats-claude-haiku-5-5-at-11x-speed.md) (2026-10-13)
- [Cognition Adds Gemini 4 Argon to Devin at 75% off](https://alphasignal.ai/news/cognition-adds-gemini-4-argon-to-devin-at-75-off.md) (2026-10-13)
- [Google DeepMind's Gemini 4 Argon Tops Vals Index With 1M-Token Window](https://alphasignal.ai/news/google-deepmind-s-gemini-4-argon-tops-vals-index-with-1m-token-window.md) (2026-10-13)
- [AutoTrust AI's JEV-35B Turns Open-Source Decisions Into 204ms API Calls](https://alphasignal.ai/news/autotrust-ai-s-jev-35b-turns-open-source-decisions-into-204ms-api-calls.md) (2026-10-13)
- [Ai2's TutorMoments Benchmark Catches AI Models Helping Students Too Soon](https://alphasignal.ai/news/ai2-s-tutormoments-benchmark-catches-ai-models-helping-students-too-soon.md) (2026-10-12)
- [Meta's Muse Wonder Beats Claude on Long-Running Agent Tasks](https://alphasignal.ai/news/meta-s-muse-wonder-beats-claude-on-long-running-agent-tasks.md) (2026-10-12)
- [Z.ai Opens Gated Access to GLM-Legal, Its 320B Legal AI](https://alphasignal.ai/news/z-ai-opens-gated-access-to-glm-legal-its-320b-legal-ai.md) (2026-10-12)
- [UkisAI Squeezes Swift 1.5's 27B Reasoning Model Into 8.42GB for Consumer GPUs](https://alphasignal.ai/news/ukisai-squeezes-swift-1-5-s-27b-reasoning-model-into-8-42gb-for-consumer-gpus.md) (2026-10-12)
- [Bartowski Packs Qwen2.5-32B Into a 9GB File for Local Machines](https://alphasignal.ai/news/bartowski-packs-qwen2-5-32b-into-a-9gb-file-for-local-machines.md) (2026-10-12)
- [Sewon Min's Preprint Shows Smaller Batches Beat Overfitting on Repeated Training Data](https://alphasignal.ai/news/sewon-min-s-preprint-shows-smaller-batches-beat-overfitting-on-repeated.md) (2026-10-11)
- [Huihui-ai Ships Uncensored Qwen3.8 Flash-Next With 176B Parameters](https://alphasignal.ai/news/huihui-ai-ships-uncensored-qwen3-8-flash-next-with-176b-parameters.md) (2026-10-11)
- [SC117 Strips Qwen3.8-Flash-Next Refusals Without Touching 1,079 Tensors](https://alphasignal.ai/news/sc117-strips-qwen3-8-flash-next-refusals-without-touching-1-079-tensors.md) (2026-10-11)
- [ETH Zurich's Apertus 1.5 Adds Image and Audio to Its Open Model Family](https://alphasignal.ai/news/eth-zurich-s-apertus-1-5-adds-image-and-audio-to-its-open-model-family.md) (2026-10-11)
- [AutoTrust AI Turns Gemma 4 Into a 45ms Calibrated Decision API](https://alphasignal.ai/news/autotrust-ai-turns-gemma-4-into-a-45ms-calibrated-decision-api.md) (2026-10-11)
- [Conway Research Squeezes Saluki's 27B Model Into 8 GB and Beats the Original](https://alphasignal.ai/news/conway-research-squeezes-saluki-s-27b-model-into-8-gb-and-beats-the-original.md) (2026-10-10)
- [Anthropic's Claude Agents Bypassed Web Controls and Submitted a Fake Police Tip](https://alphasignal.ai/news/anthropic-s-claude-agents-bypassed-web-controls-and-submitted-a-fake-police-tip.md) (2026-10-09)
- [Alibaba's Qwen2.5-Coder Hits 775K Downloads Running on 8GB GPUs](https://alphasignal.ai/news/alibaba-s-qwen2-5-coder-hits-775k-downloads-running-on-8gb-gpus.md) (2026-10-09)
- [Anthropic's Claude Haiku 5.5 Beats Every Rival at $24 per 1,000 Tasks](https://alphasignal.ai/news/anthropic-s-claude-haiku-5-5-beats-every-rival-at-24-per-1-000-tasks.md) (2026-10-09)
- [Nous Research Offers Step 5 Preview Free for a Week on Its Portal](https://alphasignal.ai/news/nous-research-offers-step-5-preview-free-for-a-week-on-its-portal.md) (2026-10-09)
- [Princeton's ReplaySSM Boosts AI Serving Speed 1.8x by Slashing Memory Traffic](https://alphasignal.ai/news/princeton-s-replayssm-boosts-ai-serving-speed-1-8x-by-slashing-memory-traffic.md) (2026-10-09)
- [Infatoshi Squeezes GLM-5.3's 753B Parameters Into 273 GiB for Multi-GPU Workstations](https://alphasignal.ai/news/infatoshi-squeezes-glm-5-3-s-753b-parameters-into-273-gib-for-multi-gpu.md) (2026-10-09)
- [Cloudflare's Clef-Omni Brings Audio and Video Into One Decision Model](https://alphasignal.ai/news/cloudflare-s-clef-omni-brings-audio-and-video-into-one-decision-model.md) (2026-10-09)
- [SemiCat Boosts Diffusion Text AI Accuracy from 4% to 50% Without Extra Training](https://alphasignal.ai/news/semicat-boosts-diffusion-text-ai-accuracy-from-4-to-50-without-extra-training.md) (2026-10-09)
- [Stanford's NCM Generates Text 31x Faster Than Diffusion Models](https://alphasignal.ai/news/stanford-s-ncm-generates-text-31x-faster-than-diffusion-models.md) (2026-10-09)
- [Leiden Drops a Free 472-Page Textbook on Agentic LLMs](https://alphasignal.ai/news/leiden-drops-a-free-472-page-textbook-on-agentic-llms.md) (2026-10-09)
- [MSAR Cuts Transformer Training Compute by 37% With Smarter Residual Connections](https://alphasignal.ai/news/msar-cuts-transformer-training-compute-by-37-with-smarter-residual-connections.md) (2026-10-09)
- [Mamba-3's Oscillatory Rotations Beat Pure Decay Beyond 2K Tokens](https://alphasignal.ai/news/mamba-3-s-oscillatory-rotations-beat-pure-decay-beyond-2k-tokens.md) (2026-10-09)
- [Liquid AI's d1-3B Makes Structured AI Decisions in 8 Milliseconds Without Generating Text](https://alphasignal.ai/news/liquid-ai-s-d1-3b-makes-structured-ai-decisions-in-8-milliseconds-without.md) (2026-10-08)
- [Anthropic's Claude Science Fills the Sky's Missing UV Third](https://alphasignal.ai/news/anthropic-s-claude-science-fills-the-sky-s-missing-uv-third.md) (2026-10-08)
- [Harvey LAB-AA Exposes AI Legal Agents Passing Under 10% of Real Tasks](https://alphasignal.ai/news/harvey-lab-aa-exposes-ai-legal-agents-passing-under-10-of-real-tasks.md) (2026-10-08)
- [OpenAI Brings 8x Faster Ultrafast Inference to GPT-6.1 Sol](https://alphasignal.ai/news/openai-brings-8x-faster-ultrafast-inference-to-gpt-6-1-sol.md) (2026-10-08)
- [Artificial Analysis Rebuilds Intelligence Index v5 With Private Coding Tests](https://alphasignal.ai/news/artificial-analysis-rebuilds-intelligence-index-v5-with-private-coding-tests.md) (2026-10-08)
- [Epoch's AI Benchmark Finds Claude Fable 5.1 Still Needs Human Review](https://alphasignal.ai/news/epoch-s-ai-benchmark-finds-claude-fable-5-1-still-needs-human-review.md) (2026-10-08)
- [Cohere Opens Ottawa Office to Chase C$240M Federal AI Deals](https://alphasignal.ai/news/cohere-opens-ottawa-office-to-chase-c-240m-federal-ai-deals.md) (2026-10-08)
- [Meta's Llama 3.3 70B Now Fits on a Single 48 GB GPU](https://alphasignal.ai/news/meta-s-llama-3-3-70b-now-fits-on-a-single-48-gb-gpu.md) (2026-10-08)
- [IBM's Tiny TSPulse Beats Models 100x Bigger Across 75 Datasets](https://alphasignal.ai/news/ibm-s-tiny-tspulse-beats-models-100x-bigger-across-75-datasets.md) (2026-10-08)
- [OpenAI Builds ChatGPT for Teens a Full College Planning Tool](https://alphasignal.ai/news/openai-builds-chatgpt-for-teens-a-full-college-planning-tool.md) (2026-10-07)
- [OpenAI's GPT-6 Turns ChatGPT Replies Into Interactive Apps for 1.2B Users](https://alphasignal.ai/news/openai-s-gpt-6-turns-chatgpt-replies-into-interactive-apps-for-1-2b-users.md) (2026-10-07)
- [Anthropic's Claude Haiku 5.5 Slashes Costs 75% and Adds Adjustable Reasoning](https://alphasignal.ai/news/anthropic-s-claude-haiku-5-5-slashes-costs-75-and-adds-adjustable-reasoning.md) (2026-10-07)
- [QuadTok Cuts Image Tokens by 10% by Spending Them Where Detail Lives](https://alphasignal.ai/news/quadtok-cuts-image-tokens-by-10-by-spending-them-where-detail-lives.md) (2026-10-07)
- [Looped transformers: three different machines](https://alphasignal.ai/news/looped-transformers-three-different-machines.md) (2026-10-07)
- [Living Models' BOTANIC-1 Pinpoints One Melon Mutation out of 2,494 Candidates](https://alphasignal.ai/news/living-models-botanic-1-pinpoints-one-melon-mutation-out-of-2-494-candidates.md) (2026-10-07)