Sakhanda Wire
NVDA $150.96 -2.89% MSFT $508.29 +0.40% GOOGL $282.44 +0.80% META $616.38 -0.40% AMZN $267.74 +3.82%

Latest news

Older stories

MarkTechPost

Perplexity brings hybrid cloud and local compute to Mac

Perplexity has launched hybrid compute for Perplexity Computer on Apple silicon Macs. The feature is available to Pro, Max, and Enterprise subscribers using macOS 15 or later, with at least 24GB of unified memory; 32GB is recommended. The local model can be installed from the Mac app with one click, without a separate runtime or API key, and local processing does not use cloud credits.

The Verge

Google seeks Hollywood licensing deals for AI training

Google has reportedly approached several of Hollywood’s largest studios about licensing copyrighted material for training its artificial-intelligence models. The proposed arrangements would involve substantial payments to the studios, while giving Google access to material that could strengthen its position in competition with other AI companies.

TechCrunch

AfterQuery reportedly reaches a $3.2 billion valuation

AI training-data startup AfterQuery has reportedly raised a funding round that values the company at $3.2 billion. The reported valuation comes five months after AfterQuery announced a $30 million Series A at a $300 million valuation in April, representing an increase of more than tenfold in less than six months.

The Verge

Anthropic launches Claude Fable 5.1 with lower pricing

Anthropic has introduced Claude Fable 5.1 and Mythos 5.1, saying the models respond to customer concerns about cost, data retention, and safeguards. The company says Fable 5.1 performs better than Fable 5 while costing about 25 percent less in typical use and up to 45 percent less for complex agentic work. The savings come partly from lower prices for cached data that has already been processed and stored.

Hugging Face

BenchMIRT separates the capabilities behind LLM benchmark scores

Researchers have introduced BenchMIRT, a method for auditing large language model benchmarks at the level of individual questions and tasks. It uses multidimensional Item Response Theory to estimate both a model’s ability and each question’s difficulty and ability to distinguish stronger from weaker models.

TechCrunch

OpenAI prepares Astra model with expanded cybersecurity safeguards

OpenAI says it plans to release Astra soon, describing it as the first large language model to reach the company’s “critical cybersecurity threshold.” According to OpenAI, Astra can identify previously unknown vulnerabilities in computer systems and exploit them without human guidance. Access to its most advanced cybersecurity functions is expected to be restricted.

TechCrunch

Google adds Android tools for motion sickness, vision and personalization

Google has announced five Android updates intended to improve accessibility, usefulness and personalization. Motion Assist is designed for people who feel motion sickness when using a phone in a moving vehicle. It places a bubble overlay on the display that responds to the vehicle’s movement, helping align visual information with physical sensations. The feature can start automatically, be added to Quick Settings, and be adjusted by changing the bubble’s shape, color and transparency. Google says Motion Assist is available on Android 17.

The Verge

OpenAI delayed Astra development after an unreleased model’s attack

OpenAI says it postponed development of Astra, an unreleased model suite, to strengthen its safety work. The company described the decision in a blog post published Tuesday, following events involving a different unreleased model. The post did not provide a new timetable for Astra or details about its planned capabilities.

Google AI

Google recaps August AI launches and product updates

Google’s August roundup included the release of Gemini 3.7 Flash, described as a more capable model for coding, software engineering and AI agents. The company said its introductory price per million tokens was half that of Gemini 3.6 Flash. Google also introduced Gemini 3.5 Transcribe, a speech-to-text model for real-time transcription in applications such as voice agents, live captions and post-call analysis.

The Decoder

Anthropic opens Claude watermark checks to approved organizations

Anthropic is introducing an API that allows approved organizations to determine whether text contains Claude’s digital watermark. Access may be requested by regulators, law-enforcement agencies, media outlets, fact-checkers, independent researchers, educational institutions and civil-society groups in the European Union. Companies may also apply when they need to verify watermarking for their own compliance. Anthropic says access will be broadened over time.

The Decoder

Anthropic releases Claude Fable 5.1 with lower costs and stronger coding

Anthropic has released Claude Fable 5.1 and Claude Mythos 5.1, describing them as its most capable models so far. The two systems share a base model but use different safety guardrails. Fable 5.1 is broadly available, while Mythos 5.1 is currently restricted to US organizations in cybersecurity and life-sciences verification programs. Both models are the first Claude releases with built-in watermarks, and Anthropic is testing an API for detecting them.

MarkTechPost

Anthropic releases Claude Fable 5.1 and Claude Mythos 5.1

Anthropic has released Claude Fable 5.1 and Claude Mythos 5.1, which use the same underlying model with different safeguard layers. Fable 5.1 is generally available as claude-fable-5-1 through the Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud and Microsoft Foundry. Mythos 5.1 remains limited to vetted US organizations participating in Project Glasswing.