Latest news, developments and reporting on Large Language Models.
OPPO and OpenKG have launched MobileMem, the first benchmark for evaluating on-device AI memory, to standardize testing and accelerate industry development.
Bairong Intelligent reported a 43% fall in H1 2026 revenue but a 52% rise in AICC revenue, with new-scenario income up 195%, as it expands outcome-based AI agents beyond finance.
Anthropic launches Fable 5.1 and Mythos 5.1, topping 8 benchmarks and cutting agentic costs by up to 45%.
Alibaba's refreshed Qwen3.8-Max claims the No.1 spot in front-end coding on CodeArena with a score of 1,691, and is priced at $5 per million tokens, outperforming pricier rivals.
World Labs, co-founded by Fei-Fei Li, released Atlas, which it calls the first multimodal world model, creating 3D worlds from single images and enabling robot training simulations.
Nvidia released DLSS 5 with neural rendering, Dyson launched the CameraJet smart toothbrush in China, and Starman Optical agreed to acquire GoPro for $285 million, among other tech news.
Google offers Hollywood billions for AI training rights; no major studio has agreed yet.
CrowdStrike unveiled SafeMind, a family of security-specific AI models built with Nvidia, and opened a Cyber Superintelligence Lab at Fal.Con, claiming faster threat detection and remediation.
Meta debuts Muse Voice Transcribe, a real-time AI model for 20+ speakers and multiple languages, now in Meta AI Mac app and API.
AfterQuery reportedly raises $3.2B valuation round, becoming Y Combinator's fastest unicorn.
Hugging Face's BenchMIRT audits LLM benchmarks at prompt level and separates safety and reasoning signals.
OpenAI said its upcoming Astra model is the first to meet its 'Critical' cybersecurity threshold, capable of hacking unknown flaws without human guidance.
A study of nearly 19,000 academic abstracts shows AI-related vocabulary surged after late 2022, and AI can self-correct to hide its origin, weakening detection tools.
A hack by OpenAI agents on Hugging Face has triggered a fight over 'AI civilizations' language and responsibility.
A nonprofit plans to launch a spacecraft to Alpha Centauri by 2029 using a trajectory discovered by an AI system from the newly launched lab Physical Superintelligence, backed by $58 million in funding.
CrowdStrike unveiled SafeMind autonomous red-teaming AI at Fal.Con 2026, saying breakout time has hit zero.
Anthropic released Claude Fable 5.1 and Mythos 5.1 on Tuesday, featuring record benchmarks, a 75% cut to cache-read prices, and a new enterprise privacy framework. The company also secured a $35B cloud deal with Lambda.
Meta has introduced Muse Voice Transcribe, a real-time audio perception model bringing multilingual streaming dictation to Mac and developers.
A key measure of AI token prices hit a record low this week as competition intensifies, squeezing AI model providers.
CNBC's Jim Cramer dismissed the FTC lawsuit against Amazon, urging investors not to sell. He also commented on Nvidia's AI deal and market concerns over rate hikes.
A CNBC/SurveyMonkey survey finds nearly a third of U.S. students have changed or considered changing target industry due to AI, with many rethinking majors and skills.
The Pentagon has launched customized ChatGPT and Grok models tailored to warfighter needs, according to TechRadar.
Phonely has launched Alma, a voice AI model trained on over 10 million phone calls, claiming faster response times and lower cost than OpenAI's GPT-4.1.
Perplexity launches Hybrid Compute for Mac, splitting AI tasks between cloud and local models to protect sensitive data and reduce costs.