August 19, 2026

AI News Brief

Daily AI news, sourced from Brave News API
← All Briefs
01

GLM-5.3 Tops CyberGym Benchmark at 84.5% as Z.ai Delays Open-Weight Release

Z.ai's GLM-5.3 scored 84.5% on the CyberGym cybersecurity benchmark, edging past Anthropic's Mythos 5 (83.8%) and OpenAI's GPT-5.6 Sol (83.6%), while delivering a 50% coding improvement over GLM-5.2. Z.ai delayed the open-weight release by two weeks after tests revealed unexpected hacking capabilities, with the model finding over 1,000 critical vulnerabilities in security testing.

02

Alibaba Releases Qwen3.8-27B Open-Weight Model Built for Laptops

Alibaba unveiled Qwen3.8-27B, an open-weight AI model designed to run on a single consumer GPU, competing directly with Meta's on-device AI push. Alibaba also released the weights of its most powerful Qwen model, sharpening the open-source competition in the on-device AI space just days before Meta's own move.

03

DeepSeek V4 Pro 0813 Benchmarks Show Near-Frontier Performance as Price Hikes Hit

DeepSeek V4 Pro 0813 ranks 9th on the World of AI Benchmark, staying within a few points of the frontier across agentic coding tests while winning outright on several categories. DeepSeek confirmed price increases for its V4-Flash and V4-Pro models, while Zhipu's GLM-5.3 took a different path by keeping the same base model and applying post-training scaling.

04

Etched Hits $21 Billion Valuation in AI Chip Race Against Nvidia

AI inference chip startup Etched raised $700 million at a $21 billion valuation, more than doubling in less than a month, with Jane Street leading the round alongside Kleiner Perkins, Sequoia, and Andreessen Horowitz. The San Jose-based company is betting specialized inference chips can bring down the cost of running AI models at scale.

05

Vals AI Raises $40M for Independent AI Benchmarking as Frontier Models Fail 52% of Finance Tasks

Vals AI raised $40 million at a $400 million valuation led by Andreessen Horowitz to expand independent AI benchmarks across coding, cybersecurity, and professional tasks. The company's evaluations found that frontier models fail 52% of real-world finance analyst tasks, highlighting the gap between vendor-reported benchmarks and actual performance.

06

Qualcomm Snapdragon X2 Beats AMD and Intel in Flagship AI PC Benchmarks

Qualcomm's Snapdragon X2 Elite Extreme won every CPU and AI benchmark against Intel's Core Ultra X9, even while running on battery in Balanced mode against Intel plugged in at Best Performance. The Snapdragon system also surpassed AMD's Ryzen AI 9 465 in Geekbench 7 by 83%, combining high compute throughput with aggressive power efficiency for premium AI laptops.