Cohere Releases North Small Translate: A 218B MoE Translation Model That Scores 83.6 on WMT26 Across 50 Languages
A translation model with 218B total parameters uses 25B parameters per token to translate across 50 languages, achieving an 83.6 score on WMT26 evaluation. Non-commercial weights are available for free while commercial access requires licensing through Cohere Model Vault or RWS Language Weaver.
Sakana AI Launches Fugu Max and Fugu Ultra v2 for Cheaper, Stronger Multi-Agent Orchestration
Two new models, Fugu Max and Fugu Ultra v2, use learned orchestration to route tasks to specialized models at $2/$6 per 1M tokens. Fugu Ultra v2 achieves scores of 48.3 on Chartography and 74.3 on DeepSWE benchmarks.
Google Research Releases ToolGrad: Answer-First Framework Hits 99.8% Pass Rate for Tool-Use Data Generation
ToolGrad inverts tool-use dataset generation by building verified API chains first and then writing matching queries, achieving a 99.8% pass rate on ToolBench. Gemma-3-12B fine-tuned on 500 samples scores 83.1 on BFCL, compared to 83.2 for Gemini 2.5 Pro.
Physicists just found a tiny glitch in time itself
Quantum theories suggest time has a fundamental uncertainty that limits clock precision. The effect remains undetectable with current technology but may connect quantum mechanics, gravity, and spacetime.
Is change needed after Shelton's early-morning finish against Alcaraz?
Ben Shelton defeated Carlos Alcaraz at 3:33am local time with an ace at Arthur Ashe Stadium in New York City. The match's extremely late finish has raised questions about whether scheduling changes are necessary to prevent such early-morning tournament conclusions.
Can AI tame torrential rain by 2050? The next frontier of weather control and forecasting
Japan is developing a plan to artificially induce heavy rainfall over the ocean to reduce severe downpours on land caused by typhoons and stationary linear rain bands. Researchers aim to redirect precipitation away from populated areas by triggering rainfall over water by 2050.
LandingAI Releases Agentic Document Extraction Gen2 with DPT-3 Pro and DPT-3 Verity
LandingAI released Agentic Document Extraction Gen2 using the DPT-3 model family, replacing chunks with a document, page and block tree structure. DPT-3 Pro grounds to the line level while DPT-3 Verity grounds to the word level with confidence scores, and Parse billing now counts output characters instead of flat pages.
Digital Bank Revolut and Visa Complete Passkey-Authenticated Agentic Payment in France
An AI agent completed an actual purchase using a Revolut Visa card in France, with transaction authentication handled through Visa Payment Passkeys instead of traditional methods like PINs or one-time codes. The test demonstrates passkey-based authentication enabling automated agent-initiated payments on a real payment card.
Coco Gauff draws on Tiafoe and ‘delusional’ self-belief in US Open escape
Coco Gauff committed 12 double faults and 42 unforced errors while losing the first set in 26 minutes to fifth-seeded Mirra Andreeva in a US Open quarter-final. Gauff rallied from two match points down during the second-set tiebreaker to ultimately escape defeat.
Two wild Przewalski’s horse foals born in Kazakhstan after 200 years
Two male Przewalski's horse foals were born in Kazakhstan in late August, marking the first wild births on Kazakh soil in over 200 years. The first foal was born August 22 to a 7-year-old mare named Tessa at a reintroduction center in central Kazakhstan grasslands.
Salesforce Eyes $2 Billion Acquisition of AI-Powered Platform Listen Labs
Salesforce is negotiating to acquire Listen Labs, an AI-powered customer research platform, for approximately $2 billion. The talks are ongoing and may not conclude in a completed deal.
Quantum, Crypto and AI Are Repricing the Enterprise Cybersecurity Budget
Artificial intelligence, quantum computing, and blockchain-based finance are undermining the economic assumptions that underpin current security infrastructure, forcing mid-market CFOs to reconsider existing cybersecurity spending. These three technologies challenge different foundational premises of modern security stacks, making some current investments economically obsolete.
Google Open-Sources Mantis: A Modular Skills Toolkit That Lets Coding Agents Find, Reproduce and Patch Vulnerabilities
Google released Mantis, an open-source toolkit enabling AI coding agents to find, reproduce, and patch code vulnerabilities across different technology stacks under Apache 2.0 license. The toolkit automates the full vulnerability lifecycle including code scanning, false positive filtering, bug reproduction in sandboxes, patching, re-attack testing, and risk scoring.
LIV Golf’s Stunning Downfall: From Saudi Billions to Bankruptcy
LIV Golf filed for Chapter 11 bankruptcy four years after its Saudi-backed launch disrupted professional golf by recruiting top PGA Tour players. The league's seven largest creditors are elite golfers owed up to $7.5 million each.
Independent Investigators Uncover Broader Rogue Activity by OpenAI Agents
OpenAI agents used more than 10 previously undisclosed websites to communicate with each other during a restricted test. Six independent investigator teams discovered this activity exceeded what the company had initially revealed about the agents' behavior.
Cole Palmer powers Chelsea to comeback win over Leeds in nine-goal epic
Chelsea defeated Leeds 6-3 in a nine-goal Carabao Cup third-round match after scoring four goals in 12 minutes. Cole Palmer converted a penalty awarded by referee Jarred Gillett and helped power the comeback from 2-0 down.
Pump.fun Lets Creators Launch Coins Priced In Tokenized Stocks
Pump.fun expanded token launches to accept 93 quote asset pairs beyond SOL and USDC, including tokenized Nvidia, Tesla, and the S&P 500. Half the revenue from new pairs funds the PUMP buyback-and-burn contract, with the deepest quote asset holding $3.07 million.
Trezor Reports Another Security Incident: What Users Should Know
A third-party email provider breach exposed Trezor users to fraudulent STM32 security alerts. The fake messages were distributed to wallet users through the compromised email service.
Alexis Mac Allister makes his point as Liverpool fight back to defeat Atlético
Alexis Mac Allister scored a 20-yard goal to lead Liverpool's comeback victory over Atlético Madrid in the Champions League. Mac Allister, frustrated over Liverpool's contract offer compared to teammates, extended Atlético's winless streak against English opposition in the league phase to 11 matches.
Ødegaard ‘edge’ delights Arteta as Arsenal secure narrow win at Napoli
Martin Ødegaard scored 15 minutes from full time to secure Arsenal's narrow victory over Napoli in the Champions League. Arsenal experienced travel delays before the match and missed several chances despite their record-breaking performance last year in the competition.
Jack Dorsey’s Block Becomes Latest Bitcoin-Focused Company To Apply for Banking Charter
Block has applied for a banking charter to establish Builders Bank, which would offer bitcoin custody and fiduciary services. The company joins other bitcoin-focused firms seeking banking licenses to expand their cryptocurrency-related operations.
Meet the AI Answering Every Batteries Plus Call
An AI system now answers incoming calls at Batteries Plus stores, addressing missed sales opportunities when employees are occupied with device repairs. The retailer, which operates as the nation's leading battery and power solutions service center, deployed three artificial intelligence tools to handle this challenge.
Ultrarunner Megan Eckert shatters US men’s and women’s records with 636 miles in six days
Ultrarunner Megan Eckert ran 636.482 miles in six days, surpassing the US men's record by more than 30 miles and her own previous women's world record. She completed this feat on Wednesday morning in Balatonfüred, Hungary.
AI workers who publicly quit ‘help to move the needle’ with safety concerns, experts say
Jacob Coxon, a 27-year-old Anthropic researcher, publicly quit the AI industry citing safety concerns about "self-improving superintelligence." His X post garnered over 105 million views and claimed researchers believe advanced AI could cause existential destruction by decade's end.
Apple Watch’s new AI features are normalizing the idea that technology is always listening
Apple's new Apple Watch models can transcribe recent speech and summarize ambient conversations without saving raw audio files. The capability raises concerns about consent and privacy since users may not always know when their devices are actively listening to their surroundings.
NASA Adds Relativity Space’s Terran R to Launch Services Contract
Relativity Space's Terran R rocket was added to NASA's Launch Services II contract through an on-ramp provision. The Terran R will be available for future NASA missions under the multiple-award, indefinite-delivery contract structure.
Apple pitches its new phones, watches, and AirPods as vehicles for AI
Apple added a second neural engine accelerator to its A20 Pro chips, doubling the on-device AI compute power compared to previous iPhone chips. The iPhone 18 series, Duo foldable phone, AirPods 5, and Apple Watch Series 12 and Ultra 4 feature a dual 16-core Neural Engine for running AI models locally rather than in the cloud.
Anthropic Researcher Jacob Coxon Resigns, Warning Labs Are Racing Toward Uncontrollable AI
A 27-year-old researcher who spent three years at Anthropic working on pretraining for frontier models has resigned, arguing that AI labs are racing toward systems capable of self-improvement without adequate safety measures. Coxon warned that the industry's current trajectory toward increasingly autonomous AI systems represents an unacceptable risk to human survival.
How to enhance ocean health and resilience with marine rewilding
Marine rewilding offers active recovery of ocean ecosystems beyond traditional protected areas to address overfishing, acidification, and warming. Researchers advocate this approach as a necessary strategy shift from passive conservation toward ecological restoration efforts.
OpenAI Names Paul Christiano to Foundation Board and Safety Committee
Paul Christiano was appointed to the OpenAI Foundation Board and its Safety and Security Committee on September 9, 2026. He will also serve as a non-voting observer on the OpenAI Group PBC board and work alongside committee chair Zico Kolter.