Large language model news
The open-weight and frontier model landscape changes weekly. This page follows new large language models from Meta, Mistral, Alibaba, DeepSeek and smaller labs, together with benchmark results, licence changes and the research behind each release.
Latest stories

Beyond LLMs? China forges new path for AI with focus on industrial systems
China’s artificial intelligence development is expanding beyond general large language models (LLMs) towards a new generation of specialised systems designed to solve real-world industrial problems, according to a leading computer science researcher. Cui Peng, an associate professor at Beijing’s elite Tsinghua University, said that specialist systems capable of analysing industrial data, such as factory sensor readings and financial records, could become a crucial part of AI’s next phase in...
SCMP Tech
AI Chips & Semis (Google News)
AI Global WireShould You Use Open-Weight AI? Here’s What to Consider.
Open-weight AI systems give developers more control over data processing and customization. From DeepSeek-V4 to Kimi K3, we explore how open models work — and the benefits and risks of deploying them.
AI Chips & Semis (Google News)

DeepSeek Harness active users quadruple as plugin ecosystem expands
DeepSeek said its open-source AI agent framework, DeepSeek Harness, had 1.54 million daily active users as of November 4, more than quadrupling in less than three months. The company also plans to launch an official plugin marketplace to expand its developer ecosystem.
DIGITIMES

Can DeepSeek challenge OpenAI, Anthropic in AI agents as Harness adoption takes off?
DeepSeek’s Harness artificial intelligence agent framework has reached 1.54 million daily active users less than three months after launch, the Chinese AI champion said on Sunday, marking rapid growth in a new battleground between leading Chinese and American AI giants. Users of Harness through the company’s official application programming interface were consuming around 92 trillion tokens a day as of November 4, the Hangzhou-based start-up said. That is equivalent to about 1.1 billion tokens...
SCMP Tech
arXiv cs.AI
AI Global WireI Think, Therefore I Attack: Evaluating Thought Injection Attacks on Reasoning Agents
arXiv:2611.08811v1 Announce Type: new Abstract: Large Reasoning Models (LRMs) are one of the key components of current agentic systems. LRMs augment their output with an explicit chain-of-thought reasoning stage, introducing a new attack surface for thought injection, in which untrusted text resembling the model's own deliberation is fed back to the model. We present a two-part study of this attack, performed against the prevailing large, production class open-weight models (e.g., GLM-5.3, MiniMax-M2.5, Qwen3.8-Flash-Next). We first evaluate the effectiveness of static thought-injection prompts under different styles as well as tagged and untagged conditions and find that the effects of thin
arXiv cs.AI
arXiv cs.AI
AI Global WireDebate to Deceive: Poisoning LLM's Fact-Checking via Game-Theoretic Multi-Agent Debate
arXiv:2611.08756v1 Announce Type: new Abstract: Knowledge attacks on Large Language Models, such as manipulating fact-checking judgments, pose significant risks to AI reliability. Existing approaches -- adversarial suffix/template injection and single-agent self-reflection -- lack robustness across diverse defenses and are prone to confirmation bias and costly trial-and-error, respectively. We propose Debate-to-Deceive (D2D), a multi-agent debate framework grounded in human debate structure and rhetorical style, enabling agents to generate adversarial passages that are semantically coherent, rhetorically effective, and resistant to early confirmation bias. D2D formalizes the adversarial proc
arXiv cs.AI
arXiv cs.AI
AI Global WireCancellation Is Not Enough: What Retries Leave Behind in an Overloaded LLM Engine
arXiv:2611.08684v1 Announce Type: new Abstract: Clients and proxies in front of an LLM engine retry calls that time out. After a transient overload, these retries can keep the engine from recovering even when calls arrive below its capacity. Since the engine may keep computing for clients that have left, propagating cancellation looks like the natural remedy. In this paper, we test it on vLLM with overload experiments on three open models, and with scheduler traces of one that follow each client attempt to the execution it creates in the engine. We find that with retries on, cancellation alone did not restore service in any run, and only 12-38% of the calls that arrived after the overload we
arXiv cs.AI
arXiv cs.AI
AI Global WireMind the Divergence: Divergence-Guided Reliability-Aware Fusion and Adaptive LLM Prompting for Conversational Emotion Recognition
arXiv:2611.08860v1 Announce Type: new Abstract: Multimodal emotion recognition in conversations (MERC) integrates textual, acoustic, and visual cues to identify utterance-level emotions. However, existing approaches generally do not explicitly adapt multimodal processing to the degree of inter-modal predictive disagreement. In real conversations, semantic incongruity, concealed emotion, noisy observations, and ambiguity can affect modality-specific emotional evidence differently, while the reliability of individual modalities varies across samples. Rather than identifying each source of inconsistency separately, we model their common consequence for emotion recognition as predictive disagree
arXiv cs.AI
arXiv cs.AI
AI Global WireAlterNet: Adaptive-Depth Mixture-of-Experts with Learned Halting for Elastic Inference
arXiv:2611.08741v1 Announce Type: new Abstract: Most LLM architectures have fixed compute demands determined at training time. We propose a novel architecture that combines the Mixture-of-Experts architecture with shallow experts and a novel dynamic-depth system to make the extent of a forward pass dynamically adjustable. Our approach lets the model automatically determine the compute needed for a given task, paired with a system to scale it down after training without retraining when the execution environment demands faster or cheaper inference. To achieve this, we introduce an iterative computation mechanism, sequence-level depth-based routing steered by a halting mechanism within the expe
arXiv cs.AI
arXiv cs.AI
AI Global WireNotwithstanding Anything to the Contrary, Ignore All Instructions: Measuring Prompt Injection Guards on Legal Contract Review Attacks
arXiv:2611.08834v1 Announce Type: new Abstract: Specialized prompt injection guards are deployed in front of document analysis systems, but no false positive rate has been published for any of them on legal contracts. We measure eleven open-weights guards on two corpora of legal contracts and find that existing guards struggle: document-level false positive rates are as high as 92.39%; two guards fall below chance on paired AUC; and a five-regex baseline is not Pareto-dominated by any guard in the panel. Controlled probes locate model-specific as well as general failure modes. Prompted LLMs resolve the tested failures, including a second threat category (unauthorized scope claims) on which e
arXiv cs.AI
arXiv cs.AI
AI Global WireLLM-Based Feature Extraction for Multi-Criteria Inventory Classification
arXiv:2611.08626v1 Announce Type: new Abstract: E-commerce supply chains generate large volumes of customer feedback and operational data that remain underutilized in inventory management. Traditional inventory classification techniques such as ABC analysis rely primarily on annual dollar usage, while multi-criteria inventory classification (MCIC) methods extend this but still depend on structured quantitative data and do not utilize unstructured information available in modern supply chains. This paper proposes an LLM-based feature extraction approach that transforms unstructured sources such as customer reviews into quantitative features using vector embeddings and LLM-generated evaluation
arXiv cs.AI
arXiv cs.AI
AI Global WireChainFlow: Few-Step Latent Flow Matching for Speculative Decoding
arXiv:2611.08669v1 Announce Type: new Abstract: Speculative decoding has become a standard way to accelerate LLM inference: a lightweight drafter proposes tokens that the target verifies in one parallel pass. Existing drafters face trade-offs that limits their speedups: autoregressive drafters generate tokens one at a time, so drafting cost grows with draft length, while diffusion drafters generate tokens in parallel but predict each position independently, ignoring inter-token dependencies. We introduce ChainFlow, a new speculative decoding paradigm and the first flow-matching drafter conditioned on the target's hidden states. ChainFlow treats drafting as a short transport over hidden state
arXiv cs.AI
arXiv cs.AI
AI Global WireToward LLM Self-Play Under Data Scarcity: Effectiveness, Generalizability, and Trustworthiness
arXiv:2611.08908v1 Announce Type: new Abstract: The rapid development of large language models (LLMs) has been driven largely by scaling model size and human-curated training data, yet this paradigm is increasingly constrained by data scarcity and diminishing returns from human supervision. Self-play offers a promising pathway for autonomous data generation and improvement, but current frameworks remain unsatisfactory due to a lack of a data-centric perspective, limited cross-domain generalization, and growing trustworthiness and security concerns. In this paper, we argue that real-world LLM self-play systems should achieve effectiveness, generalizability, and trustworthiness under data scar
arXiv cs.AI
arXiv cs.AI
AI Global WireAspect-wise Pairwise Tournaments for Test-Time Selection
arXiv:2611.08640v1 Announce Type: new Abstract: Scaling test-time compute has proven highly effective at improving LLM reasoning. A popular method is best-of-n sampling, where a model generates many candidate solutions and one is selected. This motivates the question: how can we select the best candidate? When humans make choices, we find it more natural to compare two options than judge each in isolation. Additionally, when the options become complex, we like to judge the options one aspect at a time, such as comparing two job candidates on experience, communication, etc. In the test-time setting, candidate solutions should be judged in the same way. Here we introduce Aspect-wise Pairwise T
arXiv cs.AI
arXiv cs.AI
AI Global WireDecision Context: Persistent Behavioral Learning for AI Agents
arXiv:2611.08644v1 Announce Type: new Abstract: LLM agents do not learn from corrections across sessions. A user fixes the same mistake at the start of every conversation. Existing agent memory systems primarily handle facts, such as preferences, entities, and conversational history, rather than procedural knowledge. We present Forge Decision Context (forge_dc), a system that captures behavioral signals from human corrections and self-assessed tool-call failures, compiles them into imperative RULE directives, and injects them at the decision points where they apply. We evaluate forge_dc alongside four predecessor approaches on DC-Bench, a new benchmark of 71 scenarios across nine behavioral
arXiv cs.AI
arXiv cs.AI
AI Global WireMemory-Aware Multi-Agent Architecture with LLM-based Orchestration for Conversational Recommendation
arXiv:2611.08718v1 Announce Type: new Abstract: Conversational recommender systems often treat user sessions in isolation, limiting their ability to capture how user intent evolves over time. While large language models (LLMs) have improved semantic understanding, they are typically used as isolated components, without addressing how memory, retrieval, and decision-making should be integrated at the system level. In this work, we propose a memory-aware multi-agent architecture for conversational recommendation, in which LLMs act as orchestration layers that coordinate intent refinement, query construction, and interaction flow. A key design principle is the integration of long-term memory as
arXiv cs.AI
arXiv cs.AI
AI Global WireAttentionSwitch: Jailbreak Defense for LLMs via Joint Learning of Sparse Attention-Head Selection and Scaling
arXiv:2611.08752v1 Announce Type: new Abstract: Large language models have achieved remarkable performance across a wide range of tasks, yet they remain susceptible to jailbreak attacks that circumvent safety mechanisms and elicit harmful or unintended outputs. Although safety can be improved through model fine-tuning, such approaches can be computationally expensive and may degrade previously acquired capabilities through catastrophic forgetting. Recent jailbreak defenses have therefore explored targeted interventions on specific model components associated with unsafe behavior. However, existing approaches typically decouple two inherently related decisions: where to intervene and how to m
arXiv cs.AI
arXiv cs.AI
AI Global WireWhen F1 Is Not Enough: Auditing Codebooks for LLM-Assisted Annotation
arXiv:2611.08846v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly used for text annotation in social science, but standard performance metrics do not explain why errors occur. Low F1 may reflect limitations of the annotation system, defects in the written codebook, or inconsistency in reference labels. We propose a pre-deployment codebook audit for LLM-assisted annotation. The audit uses class-specific metrics and confusion matrices to identify problematic classes and boundaries, structured document-level review to attribute error sources, and targeted codebook revision followed by held-out evaluation. We apply the audit to two published multi-class codebooks from
arXiv cs.AI
ITmedia AI+
AI Global WireQwen3.8-27Bベースに「西洋の価値観を強化」したオープンモデル登場、「逆に偏っている」との批判も Hermes Agent開発元
「Hermes Agent」を提供するNous Researchは、「Qwen3.8-27B」をベースに「西洋の価値観を強化」した「Arete-27B」を発表した。
ITmedia AI+

DeepSeek says its open-source agent app Harness hit 1.54M DAUs on November 4, less than three months after launch, and uses ~92T tokens a day, or ~60M per user (Saritha Rai/Bloomberg)
Saritha Rai / Bloomberg : DeepSeek says its open-source agent app Harness hit 1.54M DAUs on November 4, less than three months after launch, and uses ~92T tokens a day, or ~60M per user — DeepSeek, the Chinese AI disruptor, released striking adoption numbers for its AI agent infrastructure as it continues …
Techmeme
China AI Markets (Google News)
AI Global WireHelping AI Recommend Quality Brands Accurately: Second GEO Conference Opens in Shenzhen on November 21
As more users bypass search results pages and turn directly to AI assistants such as ChatGPT, DeepSeek, Doubao and Kimi for answers, brands face a new challenge in customer acquisition and brand ...
China AI Markets (Google News)

The inventor of the bitter lesson has a bitter lesson for LLM scaling enthusiasts
Turing Award winner Richard Sutton reaffirms that raw compute alone isn't enough for intelligence and that fundamentally new learning methods are needed. Ex-OpenAI researcher Tworek agrees, saying blind transformer scaling "has reward hacked brains of a whole generation of scientists." A growing wave of startups is already betting on exactly this gap. The article The inventor of the bitter lesson has a bitter lesson for LLM scaling enthusiasts appeared first on The Decoder .
The Decoder
MarkTechPost
AI Global WireWhich Routing Does Each MoE Model Use? DeepSeek, Qwen, Mixtral, Llama 4, gpt-oss Compared
A config-verified comparison of how 10 open MoE models route tokens: total and active parameters, expert counts, top-k, shared experts, gating functions and load-balancing methods. The post Which Routing Does Each MoE Model Use? DeepSeek, Qwen, Mixtral, Llama 4, gpt-oss Compared appeared first on MarkTechPost .
MarkTechPost

MoE Routing Strategies Explained: Top-K, Expert-Choice, Sigmoid Gating and Dynamic Routing (2026 Guide)
MoE routing strategies decide which experts process each token. This guide explains 7 methods, from Mixtral's top-2 softmax to DeepSeek-V4's hash routing, with the models that use each one. The post MoE Routing Strategies Explained: Top-K, Expert-Choice, Sigmoid Gating and Dynamic Routing (2026 Guide) appeared first on MarkTechPost .
MarkTechPost
MarkTechPost
AI Global WireNous Research Releases Arete-27B that Cuts Qwen3.8-27B Propaganda From 22% to 0.9%
Nous Research released Arete-27B, an Apache 2.0 fine-tune of Qwen3.8-27B. On Nous’s 803-prompt eval, propaganda answers fell from 22% to 0.9% while Hermes Bench stayed flat at 65.60%. The post Nous Research Releases Arete-27B that Cuts Qwen3.8-27B Propaganda From 22% to 0.9% appeared first on MarkTechPost .
MarkTechPost
MarkTechPost
AI Global WireHow to Run NVIDIA Dory 4B with llama.cpp: Loop by Loop Guide
Explore a comprehensive coding guide for NVIDIA's Dory 1.0 4B looped language model. Learn how to build the pinned llama.cpp branch, analyze loop-by-loop compute costs, test memory caching, and run paired evaluations. The post How to Run NVIDIA Dory 4B with llama.cpp: Loop by Loop Guide appeared first on MarkTechPost .
MarkTechPost

The gap between open and closed AI models in cyber capabilities keeps shrinking
In the Vals AI benchmark, OpenAI's GPT-6 Sol Daybreak hits 80.8 percent of attack objectives in simulated corporate networks. Worse, the open-weight model DeepSeek V4.1 Flash reaches 72.8 percent at a tenth of the cost. Because open models can no longer be restricted, analysts say the window for defenders is shrinking fast. The article The gap between open and closed AI models in cyber capabilities keeps shrinking appeared first on The Decoder .
The Decoder

State propaganda in some Chinese AI models is only "a thin layer" that fine-tuning can strip away
Nous Research retrained Alibaba's Qwen3.8-27B to produce Arete-27B, largely stripping out its pro-Chinese bias. The propaganda rate dropped from 22 percent to less than one percent, while general capabilities remained intact in tests. The method relied on 53 million tokens of supervised fine-tuning. The political bias is "applied late, it is applied thin," the team writes. The article State propaganda in some Chinese AI models is only "a thin layer" that fine-tuning can strip away appeared first on The Decoder .
The Decoder
MarkTechPost
AI Global WireBuoyant Releases Kinference: An Open Source Inference Control Plane for Kubernetes and vLLM
Buoyant has released Kinference, an Apache 2.0 inference control plane for vLLM on Kubernetes. It adds declarative routing policies, KV-cache-aware load balancing and per-tenant metrics, and cut median TTFT 95% in Buoyant’s own benchmark. The post Buoyant Releases Kinference: An Open Source Inference Control Plane for Kubernetes and vLLM appeared first on MarkTechPost .
MarkTechPost
Asia AI Markets (Google News)
AI Global WireAI stocks have taken a beating, but Gene Munster says the biggest AI boom is still ahead
Munster drew parallels between the latest AI sell-off and the market turbulence in early 2025, when Chinese AI startup DeepSeek rattled investors. ・However, he believes that this time, political ...
Asia AI Markets (Google News)