12 posts from the last 24 hours on temperature2.

Z.AI's GLM-5.3-Flash nears Opus 4.8 at one-tenth the cost

Z.AI published GLM-5.3-Flash's weights under an MIT license, naming the model it had run anonymously as Ox Alpha, and priced it at $0.15 per million input tokens.

Nvidia's $96.2B beat comes with a margin warning

Nvidia's Q2 FY2027 revenue hit $96.2 billion, up 106% year over year, but shares dipped as Wall Street fixated on thinning margins tied to rising memory costs.

Amazon shuts down Mechanical Turk after 21 years

Amazon will close AWS Mechanical Turk on September 30, 2026, ending 21 years of crowdsourced microtask labor as rivals paying $95/hour for expert AI data hit $20B+ valuations.

Qwen3.8-Flash-Next activates 6B of its 125B parameters

Alibaba's Qwen team shipped a 125B-parameter MoE model that only wakes up 6B params per token, built on a new hybrid attention design.

Signals: Vera's 88 cores, wellbeing grants, agent data

Nvidia detailed its 88-core Vera CPU at Hot Chips 2026, Anthropic put $5M behind independent AI-wellbeing research, and a new paper trains web agents on 203,000 browser trajectories.

Taiwan charges 9 over smuggled Nvidia B300 chips to China

Taiwan charged 9 people, including staff at Nvidia and Super Micro, over a ring that smuggled 74 restricted Nvidia B300 AI servers into China and got caught moving 56 more.

Bill Gates says AI has crossed its danger thresholds

Gates told MIT Technology Review AI has crossed bio, cyber, psychosocial and job-destruction thresholds, and published a companion essay warning there's no plan for the transition.

Z.AI confirms it built the Ox Alpha stealth model

Z.AI ended five days of guessing by confirming it built Ox Alpha, the free stealth model topping OpenRouter's chart, and said it will release the weights.

OpenAI's first chip Jalapeño beats Nvidia Blackwell on inference

OpenAI published its first independently-verified chip benchmarks on August 25, 2026, showing its in-house Jalapeño ASIC beating Nvidia's Blackwell and next-gen Rubin on inference work per watt.

Anthropic tells IPO investors its market tops $30 trillion

Anthropic is pitching a $30 trillion+ total addressable market and a $2 trillion valuation ahead of an IPO that could raise $100 billion, the Wall Street Journal reported.

Nvidia's Jetson Orin Nano 2 doubles inference at 78 TOPS

Nvidia's new edge robotics chip hits 78 TOPS at 40% less power, with Wing and Doosan Bobcat already building on it.

OpenAI's data center chief becomes its 13th 2026 exit

Chris Malone's exit after 17 months is OpenAI's 13th senior departure this year, hitting the team building its $500B Stargate buildout.

Written and shipped by the temperature2 pipeline. Maximum entropy, minimum filter.