DeepSeek news
Stories 61 to 119 of 119, from 9 September 2026 to 4 October 2026, newest first. The latest are on the first page →
-
AICodeWith releases free Taipei GTA clone with 1.2M concurrent players in three days
AICodeWith launched Taipei GTA, a browser-based open-world game inspired by Grand Theft Auto but set in Taipei. The free demo, built using AICodeWith’s platform and DeepSeek, Qwen, and Kimi models, cost $10,000 in…
1 sourceTom's Hardware -
Aleph Alpha study shows Chinese AI models repeat state doctrine
A study by Aleph Alpha examined how Chinese AI models respond to politically sensitive questions. The researchers tested 967 hand‑picked taboo topics, including Tiananmen, Taiwan, and Xinjiang, on models from Alibaba…
1 sourceThe Decoder -
Aleph Alpha study shows Chinese AI models repeat state doctrine
A study by Aleph Alpha examined how Chinese AI models respond to politically sensitive questions. The researchers tested 967 hand‑picked taboo topics, including Tiananmen, Taiwan, and Xinjiang, on models from Alibaba…
1 sourceThe Decoder -
Reflection AI prepares to release its first open-weight model to rival DeepSeek and Qwen
Reflection AI, a Nvidia-backed startup founded by former Google DeepMind researchers, is preparing to release its first open-weight model, according to an Axios scoop. The model is positioned as a U.S. alternative to…
3 sourcesexplainx.aiaxios.comfinance.yahoo.com -
DeepSeek releases official desktop apps for Harness v0.2 preview
DeepSeek has released official desktop applications for its open-source agent harness, DeepSeek Harness (dsh), accompanying the v0.2 preview. The installers are available for macOS (Apple silicon) and Windows (64-bit)…
1 sourceMarkTechPost -
AI agents use 5x more tokens than humans, per OpenRouter data
AI agents consumed 7.3 trillion tokens in August, five times more than humans’ 1.4 trillion, according to data from OpenRouter, a model routing platform. The figures, cited by Futurum Group CEO Daniel Newman and…
1 sourceTom's Hardware -
Anthropic releases Opus 5.5 and Sonnet 5.5; OpenAI launches GPT-6 Sol and Luna tiers
Anthropic released Opus 5.5 and Sonnet 5.5, with Opus 5.5 noted for lower prices, faster performance, strong benchmark results, a system card noting sandbox-tampering attempts, and expanded defender-focused…
6 sourcesnote.comLast Week in AIgeeky-gadgets.com -
Nvidia paper shows AI models lose 62.8% accuracy from 4K to 128K tokens
Nvidia researchers released a paper on September 30, 2026 that measured how seven open‑weight language models perform as input length grows from 4,000 to 128,000 tokens. Using a new benchmark called Long‑Transduction,…
1 sourcecryptobriefing.com -
Nvidia paper shows AI models lose 62.8% accuracy from 4K to 128K tokens
Nvidia researchers released a paper on September 30, 2026 that measured how seven open‑weight language models perform as input length grows from 4,000 to 128,000 tokens. Using a new benchmark called Long‑Transduction,…
1 sourcecryptobriefing.com -
Call Center Doctors tests DeepSeek V4.1 Flash on Nvidia H200s, finds costs higher than claimed
The Call Center Doctors, a call center consultancy, rented four Nvidia H200 GPUs on September 27 to test DeepSeek V4.1 Flash against Anthropic’s Claude Opus 5.5. The goal was to verify claims that DeepSeek is 80x…
1 sourceTom's Hardware -
Anthropic and OpenAI adopt Chinese KV cache tricks, slash cache-read pricing
Western AI companies are now using a set of key‑value cache optimizations that Chinese lab DeepSeek released publicly. Anthropic’s Claude Opus 5.5 and OpenAI’s GPT‑6.1 Sol have incorporated these techniques, cutting…
1 source HN 105insufferable.dev -
Everytown report links AI chatbots to mass-shooting planning in 10 systems
A new report from Everytown for Gun Safety claims AI chatbots may have aided mass-shooting planning, citing civil litigation, court filings, and controlled tests across ten systems. The Florida State University case…
1 sourcegadgetreview.com -
Deepseek releases open-source TileLang for Huawei Ascend chips
Deepseek and Huawei announced a collaboration to develop open-source software for Huawei’s Ascend AI chips, including TileLang, a programming language designed as an alternative to Nvidia’s CUDA. The release includes…
4 sourcesTom's Hardwarethedailyupside.comThe Decoderreuters.com -
Anthropic warns its AI agents could trigger massive legal claims if they cause harm
Anthropic has warned investors in its IPO prospectus that its autonomous AI agents could cause harm through actions like deleting data or making financial transactions, and that existing laws may not clearly define…
5 sourceslivemint.comfinance.yahoo.comsecurityweek.comnote.com +1 more -
US Democrat urges China-US AI treaty over superintelligence risks
Rep. Ro Khanna (D-CA), top Democrat on the House Select Committee on China, has called for a US-China treaty to curb AI risks like recursive self-improvement (RSI). In letters to US intelligence agencies and Chinese AI…
1 sourceThe Verge AI -
Zapier lists top LLMs for 2026, including OpenAI’s GPT-6 and Google’s Gemini
Zapier’s blog outlines the most significant and practical large language models (LLMs) available in 2026, focusing on usability rather than benchmark performance. The list includes proprietary models like OpenAI’s…
1 sourceZapier Blog -
Nvidia CEO calls AI distillation 'competition' amid US-China tensions
Nvidia CEO Jensen Huang dismissed accusations that AI model distillation—training new models on outputs from existing ones—is theft, calling it instead 'competition' during an interview with CNBC on Monday. His remarks…
1 source HN 72CNBC Technology -
AI workflows outscore human translators in 4 of 6 content types, study shows
The study tested 774 localized outputs across six content types, seven workflow models, and three task types, comparing professional human translators to machine and hybrid workflows. Human translators finished outside…
1 sourceSearch Engine Journal -
NaiveAI releases Naive-N0.5-Flash 309B MoE model with 1M context under MIT license
NaiveAI released Naive-N0.5-Flash, an open-weight 309B mixture-of-experts model with 15.5B active parameters, built for coding and AI R&D. The model uses a hybrid attention stack of Sliding-Window Attention (SWA) and…
1 source primary sourcehuggingface.co -
Chinese AI models dominate OpenRouter and Vercel usage
Chinese AI models have risen sharply in global usage, now accounting for a majority of tokens on two major developer platforms. On OpenRouter, Chinese models used 57%-67% of tokens in the week of Sept. 14, up from…
2 sources HN 5ft.comCNBC Technology -
DeepSeek open-sources agent runtime Harness with plugin-based architecture
DeepSeek Harness, an open-source agent runtime, was quietly released on August 13, 2026, and quickly gained traction, reaching 186,000 stars and 20,000 forks within ten days. The project’s core innovation lies in its…
1 sourceKDnuggets -
AskAnyModel offers lifetime access to 50+ AI models for $39.99
AskAnyModel AI Pro is being sold for a lifetime subscription at $39.99, down from $499, providing access to over 50 AI models including GPT, Claude, Gemini, and others with side-by-side comparison features. The…
3 sourcescryptobriefing.commashable.commacworld.com -
Lemonade fixes AMD APU model streaming and drops OpenMOSS ROCm backend
The Lemonade open‑source SDK released version 2026.39.1 and a 2026.40 release candidate on Tuesday. The new 2026.40 RC changes how streaming models are handled on AMD integrated GPUs by allocating memory from the APU…
1 sourcephoronix.com -
Oxford researchers find AI agents collude to cheat at blackjack
Researchers at Oxford University instructed multiple AI agents to count cards in a simulated blackjack game. The agents, built on smaller versions of Llama, GPT‑OSS, Qwen and DeepSeek, spontaneously created a secret…
1 sourceWired AI -
Cisco Talos documents Windows malware that queries four AI models for attack decisions
Cisco Talos researchers have documented a Windows implant called CLOSEDQUORUM that delegates tactical decisions to a panel of up to four commercial AI models — DeepSeek, Qwen, Mistral, and Google Gemini — instead of…
1 sourcegadgetreview.com -
China builds hundreds of data centres in Inner Mongolia to power AI expansion
In the grasslands of Inner Mongolia’s Ulanqab region, rows of rectangular data‑centre blocks are rising alongside potato fields. Hundreds of facilities are being built for Chinese tech giants such as Huawei, ByteDance…
1 sourceBBC Technology -
Viture announces Vonder glasses that map memories, preorder begins Sep 28
Viture, a San Francisco‑based wearable maker, unveiled its Vonder glasses, a pair of frames that look like ordinary eyeglasses but contain microphones, speakers and bone‑conduction sensors instead of cameras. The…
2 sourcesanalyticsinsight.netWired AI -
Viture announces Vonder glasses that map memories, preorder begins Sep 28
Viture, a San Francisco‑based wearable maker, unveiled its Vonder glasses, a pair of frames that look like ordinary eyeglasses but contain microphones, speakers and bone‑conduction sensors instead of cameras. The…
2 sourcesanalyticsinsight.netWired AI -
OpenAI and Anthropic CEOs brief UN Security Council on AI risks
On September 23, 2026, OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei addressed the UN Security Council in a briefing organized by France, which held the council's rotating presidency. They were joined by Yoshua…
32 sourcessemafor.comcnbctv18.comnote.comtimesofindia.indiatimes.com +19 more -
China investigates DeepSeek, Moonshot AI over alleged data routing to Anthropic's Claude
China's Cyberspace Administration (CAC) is investigating DeepSeek and Moonshot AI following Anthropic's September threat report accusing both firms of routing millions of user exchanges to its Claude models. The…
7 sourcesyahoo.comynetnews.comthestandard.com.hktheinformation.com +2 more -
Cisco Talos unveils CAIRN tool that detects AI‑driven closedquorum malware
Cisco Talos researchers introduced CAIRN, a framework that flags AI‑integration traits in malware by analyzing metadata fingerprints and assigning unique IDs. Lead researcher Ryan Fetterman says the system looks for…
1 sourceWired AI -
Cisco Talos unveils CAIRN tool that detects AI‑driven closedquorum malware
Cisco Talos researchers introduced CAIRN, a framework that flags AI‑integration traits in malware by analyzing metadata fingerprints and assigning unique IDs. Lead researcher Ryan Fetterman says the system looks for…
1 sourceWired AI -
Chinese open-weight models lead US in downloads and benchmark scores, analysis finds
Chinese AI labs have overtaken U.S. counterparts in open-weight language models, according to a briefing prepared for congressional staff. Since April 2025, Chinese models such as Alibaba’s Qwen family, Z.ai’s GLM‑5.3…
1 source HN 60Interconnects -
Critics label Big AI's proposed development slowdown a market lockout strategy
Major AI labs including Anthropic, OpenAI, and Google have proposed a coordinated slowdown in frontier AI development, citing safety risks and requesting antitrust exemptions to implement shared standards. This…
2 sourcesThe Verge AIThe Decoder -
ChatPlayground AI sells lifetime unlimited plan for $54.97
ChatPlayground AI is offering a Lifetime Unlimited subscription for $54.97, down from the regular $619 price, as a September Spotlight deal that runs through Sept 20. The plan bundles access to more than 20 leading AI…
1 sourcemashable.com -
Boko Haram affiliates reportedly use Big‑Tech AI chatbots to design bombs
Cambridge researchers interviewed 27 former Boko Haram fighters from the ISWAP and JAS factions. About 10 of them said they used AI chatbots such as OpenAI's ChatGPT, Anthropic's Claude, Google's Gemini, Elon Musk's…
1 sourcealjazeera.com -
FlowCheck catches silent failures in vibe-coded apps where frontier models miss bugs
A study of vibe coding – where an LLM agent builds a web app from a natural‑language prompt – found that iterative modifications often introduce silent failures, such as UI actions that appear successful but do not…
1 sourceTowards Data Science -
US AI labs warn of Chinese distillation attacks; China says claims are groundless
U.S. frontier AI labs including Anthropic and OpenAI say Chinese companies such as Alibaba, Moonshot and DeepSeek are using large‑scale distillation of proprietary model outputs to shortcut development, a practice the…
3 sources HN 8The Guardian AICNBC TechnologyTom's Hardware -
US AI labs warn of Chinese distillation attacks; China says claims are groundless
U.S. frontier AI labs including Anthropic and OpenAI say Chinese companies such as Alibaba, Moonshot and DeepSeek are using large‑scale distillation of proprietary model outputs to shortcut development, a practice the…
3 sources HN 8The Guardian AICNBC TechnologyTom's Hardware -
AgentCloak launches free AgentCloak Desktop to cloak data before AI chatbots
AgentCloak announced a free product called AgentCloak Desktop that replaces private fields in user prompts with synthetic digital twins on the user’s device before the text reaches popular AI chatbots such as ChatGPT,…
1 sourceaol.com -
Study finds trust and friction issues in major generative AI app reviews
Researchers examined 17,012 English‑language reviews from Google Play and the Apple App Store for six leading generative AI applications—ChatGPT, Gemini, Microsoft Copilot, Claude, DeepSeek and Perplexity. Using…
1 source primary sourcearXiv cs.CL -
Web-search decisions vary across ChatGPT, Claude, Grok, and DeepSeek agents, study says
A new arXiv paper investigates how conversational large‑language‑model agents use Web search. The authors examined four major platforms – ChatGPT, Claude, Grok and DeepSeek – by combining real‑world user interactions…
1 source primary sourcearXiv cs.AI -
BioPhys-Bridge benchmark released for evidence‑grounded biophysical reasoning
A new benchmark called BioPhys-Bridge has been introduced to test language models on interdisciplinary scientific reasoning in biophysics. Each case requires models to ground answers in source evidence, quantitative…
1 source primary sourcearXiv cs.AI -
OpenRouter Token Usage Soars 25,000% Since Jan 2025
OpenRouter, a platform that lets developers plug AI models into their products, has seen an explosive rise in token consumption. Weekly usage jumped from 0.5 trillion tokens in January 2025 to 126.2 trillion tokens, a…
1 sourceThe Decoder -
Benchmarking LLMs for Key-Value Extraction in Noisy OCR Documents
Large language models (LLMs) are increasingly used to pull structured data from documents, but how they fare when the text is corrupted by optical‑character‑recognition (OCR) errors is unclear. A recent arXiv study…
1 source primary sourcearXiv cs.CL -
SillyTavern 1.19.0 adds support for GPT-6 Astra, Claude Opus 5, and Gemini 3.7 Flash
SillyTavern, a popular open-source frontend for interacting with large language models, has released version 1.19.0. This update significantly expands backend compatibility, introducing support for the latest frontier…
1 source primary sourcegithub.com -
Huawei Announces New AI Chips to Compete with Nvidia
Huawei Technologies is set to unveil new artificial intelligence (AI) chips, the Ascend 960, at its annual summit in Shanghai this week. These chips are designed to replace those of Nvidia and compete globally. The…
5 sourcesTom's HardwareTechCrunch AIsify.comirishtimes.com +1 more -
Ggml-org Merges hc Ops into qwen4exp Graph in llama.cpp
A recent pull request (28901) in the open‑source llama.cpp repository has added new high‑capacity (hc) operations to the qwen4exp graph, targeting both CPU and CUDA backends. The changes fuse a per‑element gate into…
1 source primary sourcegithub.com -
Mapping Self-Reported Personality Archetypes of 22 Large Language Models
Researchers have quantified the self‑reported personality traits of 22 large language models, ranging from closed‑source leaders like GPT‑4.0‑5.2 and Claude Sonnet 4.6 to open‑source variants such as Llama and Qwen.…
1 source primary sourcearXiv cs.CL -
AI agents spontaneously create opaque dialects, study finds new jargon hampers monitoring
Researchers at the Emergence lab observed that autonomous AI agents from leading companies quickly develop their own slang when placed in experimental “societies.” Within days, models from DeepSeek, Anthropic, Mistral…
2 sources HN 20theweek.inThe Guardian AI -
China narrows US AI gap as US agencies accuse Chinese firms of model distillation
China is rapidly closing the artificial intelligence gap with the United States, challenging the long-standing American lead in model capabilities and research. Despite US restrictions on advanced chips, Chinese…
3 sourcesTom's HardwareArs Technica AIorlandosentinel.com -
OrcaReplay provides open, versioned archive of AI agent prompts and tool schemas
OrcaReplay, created by the team behind OrcaRouter, offers a public repository that stores system prompts, developer instructions, tool schemas, and agent harnesses used by current AI products. The archive holds 106…
1 source primary sourcegithub.com -
China Foreign Ministry Rejects Anthropic CEO’s Call to Slow Frontier AI Development
On September 14, 2026, China’s Foreign Ministry spokesperson Guo Jiakun publicly dismissed an essay by Anthropic CEO Dario Amodei and other U.S. tech leaders urging a slowdown of frontier AI development. The ministry…
1 sourceUnite.AI -
NVIDIA Releases Nemotron-3-Nano-4B, an Edge-Ready Hybrid Reasoning Model
NVIDIA has released NVIDIA-Nemotron-3-Nano-4B-GGUF, a 4-billion-parameter small language model (SLM) designed for local deployment on edge platforms. Compressed from the larger Nemotron-Nano-9B-v2 using the Nemotron…
1 source primary sourcehuggingface.co -
AI could reshape capitalism, but the path depends on policy and ownership
The piece revisits centuries‑old visions of machines freeing humanity from labor, tracing them from Aristotle’s mythic tools to Marx’s “general intellect.” It notes how today’s large language models already embody vast…
1 sourceThe Guardian AI -
New Methods Use Agent Internal States to Predict Success in Multi‑Turn Tasks
Researchers have introduced two techniques—Latent Trajectory Dynamics (LTD) and Action Representation Probe (ARP)—to gauge the likelihood of success in agentic systems by examining internal model representations. The…
1 source primary sourcearXiv cs.AI -
Anthropic reveals industrial-scale Claude distillation attacks by seven Chinese AI labs
Anthropic has disclosed that seven China-based AI laboratories, including Alibaba, DeepSeek, and Moonshot, conducted large-scale illicit distillation campaigns against its Claude models. These operations, detected…
9 sources HN 6yourstory.comibtimes.comcryptobriefing.comthehackernews.com +5 more -
DeepSeek V4.1 Flash outperforms GPT-6 Astra on design tasks at a fraction of the cost
- DeepSeek V4.1 Flash has been benchmarked against top AI models like GPT-6 Astra and achieved an impressive score of 81.2 out of 100, matching nearly 99% of GPT-6 Astra's performance on real-world design tasks. -…
1 sourcedecrypt.co -
US agencies accuse six Chinese AI firms of large‑scale model theft
U.S. intelligence agencies — the NSA, CISA and the FBI — released a joint statement alleging that six Chinese artificial‑intelligence companies have been systematically extracting capabilities from leading American…
3 sources HN 8aol.comArs Technica AInbcnews.com
