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

TechCrunch

Google makes visible watermarks optional for AI-generated media

Google will let users remove the visible watermark from content created with its artificial intelligence tools, including images, videos and songs. The option applies to the Nano Banana, Omni and Lyria models, according to Josh Woodward, Google’s vice president for Gemini.

TechCrunch

Meta releases Glimmer amid Zuckerberg’s call for AI access

Meta released Glimmer this week as an open-weight AI model that users can download and run on their own hardware. The launch contrasts with Muse Spark, described as the company’s more powerful model, which remains available only through Meta’s APIs. The release therefore presents two different approaches within the company: one model is distributed for local use, while the other remains controlled through an online service.

TechCrunch

Kog seeks faster AI inference from conventional GPUs

French startup Kog is developing software intended to increase the speed of AI inference on standard data-center GPUs, including AMD MI300X and NVIDIA H200 systems. A technology preview using its open-source Laneformer 2B model reached about 3,000 tokens per second per request. The model has roughly 2 billion parameters, so Kog still needs to demonstrate that its method works with larger language models.

Ars Technica

OpenAI and Anthropic cut prices as Chinese rivals advance

OpenAI and Anthropic have reduced prices for mid-tier artificial intelligence models as Chinese competitors put pressure on US labs. OpenAI lowered GPT-5.6 Luna’s cost from $1 to $0.20 per million input tokens and from $6 to $1.20 per million output tokens. Anthropic introduced Opus 5 at $5 per million input tokens and $25 per million output tokens, half the rates of its Fable 5 model. It also cancelled a planned September price increase for Sonnet 5.

The Decoder

OpenAI previews Ultrafast GPT-5.6 Sol mode at 750 tokens per second

OpenAI is previewing Ultrafast, a new inference mode for its GPT-5.6 Sol model that can generate up to 750 output tokens per second. The acceleration is provided by Cerebras under a $10 billion partnership with OpenAI. The mode is initially available through the OpenAI API, limited to selected customers, with wider access planned as capacity increases.

TechCrunch

Natural gas price risks grow for AI data center operators

Hyperscalers including Amazon, Google, Meta and Microsoft are turning to natural gas to supply power for data centers supporting their AI expansion. A report from energy research firm Noreva says prices could reach more than $10 per million BTUs at some U.S. delivery hubs in the coming years, compared with roughly $2 to $4.50 today. Henry Hub in Louisiana is currently priced at just under $3.

TechCrunch

Meta releases open-weight Glimmer alongside Zuckerberg’s AI manifesto

Meta has released Glimmer, an open-weight artificial intelligence model that can be downloaded and operated on users’ own hardware. The launch differs from Muse Spark, a more powerful Meta model that remains available only through the company’s APIs. The release coincided with a 6,500-word letter from Mark Zuckerberg arguing that artificial intelligence should be “for everyone,” rather than controlled by a small number of laboratories. The discussion also noted qualifications surrounding that vision.

AI News

Samsung presents AI models for wearable health signals

Samsung Research America has presented two foundation models designed to learn from biosignals collected by wearable devices. The research was discussed during the Health Forum at Galaxy Unpacked in July 2026, as part of the company’s Connected Care vision. The models are intended to support analysis of heart activity, sleep and physical activity, including through consumer health experiences.

The Decoder

Claude Code creates 388 maintenance pull requests for Anthropic

Anthropic is testing whether Claude Code can perform routine maintenance on the company’s own software. For the last few weeks, the system has run daily tasks across Anthropic’s iOS, Android, desktop and web applications, as well as its CLI and Agent SDK. The work is coordinated through Tag and a dedicated Slack channel, according to Boris Cherny, the Anthropic engineer who created Claude Code.

The Decoder

Zhipu AI releases GLM-5.3 for coding and cybersecurity

Chinese AI startup Zhipu AI released GLM-5.3 on August 14, 2026. The model uses the same base as GLM-5.2, while the company attributes its improvements entirely to extended post-training. Based on its own benchmarks, Zhipu says GLM-5.3 is the strongest open-weights coding model, with its largest gains appearing in agent-based tasks.

AI News

Google Health Coach to integrate Abbott’s Lingo glucose data

Abbott and Google have agreed to connect data from Abbott’s Lingo continuous glucose monitor with the Google Health app under a multiyear arrangement. Users are expected to see glucose trends alongside information about activity, sleep and other wellness measures. Google Health Coach, built with Gemini, will use the data with other health information to generate suggestions about nutrition, exercise, sleep and recovery. The service requires a Google Health Premium subscription.

The Verge

Apple reportedly develops China-focused AI model with Alibaba

Apple has reportedly trained a large language model tailored to the Chinese market in cooperation with Alibaba, according to Reuters. The report cites three unnamed people familiar with the work. The partnership is an unusual collaboration between US and Chinese technology companies amid increasing tensions between Beijing and Washington over artificial intelligence.