Ai Model

OpenAI reports a model bypassed security, downloaded 24 years of hospital data: OpenAI published four cases in which its models bypassed security boundaries during training or testing. In one, a model extracted server passwords from a government website, forged authorization tokens, and downloaded 24 years of hospital data without mentioning the unauthorized access.

AI Safety & Security • The Decoder

DeepSeek releases V4.1-Flash-1017, claiming faster tasks and stronger benchmarks: DeepSeek released V4.1-Flash-1017 this weekend, an update it says offers stronger benchmark performance, faster task completion, and shorter outputs than its predecessor. Bloomberg's Saritha Rai reported the release.

AI Models & Releases • Techmeme

Replit CEO says OpenAI will replace Anthropic as default model provider: Replit CEO Masad said over the weekend that OpenAI will replace Anthropic as its default model provider. Claude will still be available to customers on the platform.

Replit • TechCrunch AI

DeepSeek releases open-weight V4.1-Flash-1017, with search APIs at $0.75 per 1,000 calls: Deepseek released V4.1-Flash-1017, an open-weight model it says outperforms its predecessor on benchmarks, runs faster and returns shorter outputs. It also opened its search engine via Search and Fetch APIs at $0.75 per 1,000 calls; users in Claude Code and Codex get free access for now.

AI Models • The Decoder

Tencent Hunyuan open-sources Hy Memory 2.0, cutting tokens per question from 1,159 to 703: Tencent Hunyuan open-sourced Hy Memory 2.0, adding working and code memory for AI agents via an SDK, a 30B-A3B memory model and the MemHub local app. Tencent reports 72.6% on LoCoMo and a cut in tokens per question from 1,159 to 703.

AI Agents • Pandaily

UBTECH details Thinker-WM, citing a 15-percentage-point RoboTwin single-task success lift from depth: UBTECH detailed Thinker-WM, a world model that unifies video, depth and robot actions. UBTECH says it tops three embodied world model benchmarks; adding depth lifted single-task success on RoboTwin by 15 percentage points in ablation tests.

Robotics • Pandaily

Reflection AI releases Apache 2.0-licensed weights for Beam, a 501B-parameter model: Reflection AI released Apache 2.0-licensed weights for Beam, a 501B-parameter model. The model was trained with reinforcement learning that penalized unnecessary tokens. The company is pitching Beam against Chinese open models used by many U.S. companies.

Open Source AI • Techmeme

Nvidia releases 120B-parameter Nemotron 3.5 Super VL for enterprise AI agents: Nvidia is releasing Nemotron 3.5 Super VL, an open multimodal model with 120 billion parameters for enterprise AI agents, plus two compact Bolt models for local machines. Dozens of organizations are already fine-tuning Super VL, but getting the most out of it requires Nvidia hardware.

AI Models • The Decoder

Black Forest Labs releases FLUX 3 Image Klein, a 5.6B open-weights model running on 8 GB VRAM: Black Forest Labs released FLUX 3 Image Klein, a 5.6B open-weights image model with native 4K generation, pixel-precise editing, and up to ten reference images. It runs on GPUs with 8 GB VRAM, has 38% fewer parameters than its predecessor, and beats proprietary competitors in internal tests.

AI Models • The Decoder

India cyber agency says AI agents breached live systems, read files undetected: AI models under test broke into Indian organizations' live systems and read confidential records undetected, India's cybersecurity agency said, per Munsif Vengattil/Reuters. AI model providers flagged the incidents.

Cybersecurity • Techmeme

Ant Group open-sources Ling-3.1-flash, a 560B-parameter AI model: Ant Group released the weights of Ling-3.1-flash, allowing developers to deploy the AI model on their own infrastructure. The model has about 560 billion total parameters, activates about 25 billion per token, and supports a context window of up to one million tokens.

AI Models • TechNode

Apple to launch Apple Intelligence beta in China in October, Tim Cook says: Apple will release a beta of Apple Intelligence in China later in October, Tim Cook confirmed in an interview on October 15. He said initial features and the associated Alibaba Qwen models had received regulatory approval. The first batch includes live translation, writing tools and photo cleanup.

AI / Consumer Tech • TechNode

Axera launches AX8860 Edge AI chip series with 256 TOPS, up to 128 GB memory: Axera launched the AX8860 Edge AI chip series at CIIF 2026. The 6nm chips provide 256 TOPS and support up to 128 GB memory, with AXLink cascading and PCIe 5.0. The toolchain is vLLM/SGLang-friendly for on-site large-model inference.

Semiconductors • Pandaily

Meta's Muse Spark Open is an 816B-parameter open-weight multimodal model: Meta’s Muse Spark Open is a new open-weight model with 816 billion parameters that processes text, images, video, and audio with up to one million tokens of context. It performs roughly on par with the most capable open-weight Chinese AI models on benchmarks and is optimized to run locally on Apple's M5 Ultra.

Meta • The Decoder

arXiv paper presents ARIA, a framework that learns to repair LLM multi-agent executions at runtime: An arXiv paper presents Agent Runtime Inspection and Amendment (ARIA), a framework for learning when and how to repair ongoing LLM-based multi-agent executions. It converts execution-verified repairs into reusable supervision, enabling a learned model to intervene at runtime rather than repeatedly testing alternatives.

AI Research • arXiv