GLM-5.3 hits the API at $1.4/$4.4 per million tokens
Developers can now call z.ai's GLM‑5.3 via API at $1.40/$4.40 per million tokens, expanding low‑cost open‑source model use.
Daily edition
OpenAI halts Astra training as safety concerns rise, while Anthropic's Claude shows protein design promise
Daily edition
68 of 69 feeds responded
288 distinct stories
16 previously issued, held back
25 worth your time
Collection and selection
288 candidate events; 58 shortlisted stories; 43 with article text available; 25 selected. Counts describe different stages; clustering, validation and refills can change the shortlist.
Active filters: none. Sort: Ranked.
All 25 stories, in ranked orderMore filters
Anthropic says any lab can now let a language model agent run the whole protein design stack
Anthropic claims Claude agents can autonomously run protein design pipelines, achieving up to 35 % hit rates versus typical 10‑15 %.
Cerebras's Next Generation CS-4: Fast Just Got Faster
Cerebras announces its next‑gen CS‑4 system, delivering higher throughput that speeds up large‑model training and inference.
Flock Has a Powerful New AI Tool for Police. We Got Its Code
Flock Safety’s new AI can identify drivers and trace vehicle movements, linking plates to personal data for police investigations.
BREAKING: OpenAI’s unraveling has begun
Gary Marcus reports OpenAI’s trust and IPO prospects are faltering, with a worsening burn rate signaling heightened business risk.
vLLM Agentic Optimizations
vLLM adds agentic optimizations in InferenceX, cutting latency for LLM‑driven agents and lowering compute costs.
Multi-Vector (Late Interaction) Embedding Models with Sentence Transformers
Hugging Face’s Multi‑Vector encoder lets users load ColBERT‑style checkpoints for token‑level retrieval, boosting accuracy despite larger indexes.
Mojo🔥 is now open source
Mojo 1.0’s open‑source compiler under Apache 2 enables Python‑style GPU programming, easing high‑performance AI development.
How Much Memory Does Your Agent Actually Need?
Hugging Face shows agentic memory must be tuned per model tier, with strong models needing full guidelines and weaker ones a compact core.
AI Rebuilds Consumer Markets: ByteDance Large Models Unlock New Consumption Increment
ByteDance’s Doubao agents now generate competitor briefings and synthesize physical‑world training data, expanding AI‑driven enterprise workflows.
Alipay's Abao Agent Goes Into Cars and Phones: Orders Queue Into Next Year as Agent Commerce Booms
Alipay launches Abao, a full‑stack super agent for cars and phones, forecasting rapid growth of agent‑based commerce.
OpenAI Overhauls Safety Protocols After Its AI Agents Went Rogue
OpenAI halted significant Astra training runs and added stricter monitoring and alignment safeguards after detecting critical cyber capabilities.
How AI Coding Agents Can Unlock Materials Simulation with NVIDIA ALCHEMI Toolkit
NVIDIA’s ALCHEMI Toolkit offers PyTorch‑native GPU blocks, letting AI coding agents streamline atomistic material simulations.
Anthropic Risk Report: August 2026
Anthropic’s August risk report discloses extensive new safety data, including details on its leading ‘Model 2’, aiding practitioner awareness.
Memory prices up 500% in 12 months
Memory prices have surged 500 % in a year, forcing AI teams to reassess hardware budgets and model scaling.
Autonomous de novo protein binder design with Claude
Anthropic demonstrates Claude can autonomously design de novo protein binders, streamlining early drug discovery.
AI labs are failing to keep their own systems in check
Guidelight finds major AI labs, including Anthropic and OpenAI, score poorly on internal safety practices, especially prevention and containment.
Debate Training Reduces Reward Hacking in RLAIF
Research shows debate‑training opponents curtail reward‑hacking of LLM judges in RLAIF, enhancing alignment robustness.
AgentX Industry Impact: Optimizations for Agentic Workloads
AgentX introduces workload optimizations that boost efficiency of agentic LLM deployments on InferenceX.
Grok Bot for teams and enterprises
xAI releases Grok Bot, an enterprise‑focused AI assistant designed to streamline team workflows.
Block’s new Apache 2.0 agent workspace Berd works across models and harnesses, stores conversation history locally
Block open‑sources Berd, a desktop workspace that runs across models and stores agent conversation history locally.
How Claude is accelerating protein design and analytical chemistry
Anthropic reports Claude accelerates protein design and analytical chemistry workflows, cutting development cycles for researchers.
China's Optical Modules, PCBs, and Domestic AI Chips: A Day in the Market
Analysts note China’s AI‑chip sector remains resilient, with Biren Technology gaining GPU wins despite potential US module bans.
When a model reads a drug's class from its name—not its knowledge
Study finds Olmo 3 often guesses drug classes from name suffixes, exposing a shortcut that can mislead health‑related queries.
🫧 Is AI a bubble yet? Our five gauges say no
Exponential View’s analysis shows AI revenues hit $126 B and demand stays high, suggesting the sector isn’t in a bubble.
No stories match. Active filters.