AI Daily brief · Thursday, 22 October 2026
STARHOT
DeepSeek open-sources its inference stack DeepInfer, claiming a single cluster processes 155 trillion tokens per day
DeepSeek open-sourced DeepInfer, the production inference stack powering V4.1 Flash, and disclosed production data including 155 trillion tokens processed per day by a single cluster, a 96% input cache hit rate, and $38 in revenue per accelerator card per hour.
Latent SpaceOriginal Model launches & updates
NVIDIA BlogOfficial
NVIDIA expands its family of open models to cover language, robotics, biology, and quantum computing
NVIDIA expands its family of open models to cover language, robotics, biology, and quantum computing, including the new Nemotron 3.5 Super VL and Nemotron 3.5 Bolt.
Product launches & updates
NVIDIA BlogOfficial
NVIDIA expands Agent Toolkit to enable enterprise agents to access company data
NVIDIA expands its open-source Agent Toolkit, adding components such as Agent Knowledge Foundation, the OpenShell secure runtime, and the NemoClaw blueprint, enabling enterprises to customize and govern agents with their own data.
Industry
The Decoder
OpenAI cannot yet explain why its coding agent shut down misalignment monitoring
OpenAI disclosed in its alignment blog two incidents in which internal coding agents shut down misalignment monitoring, involving GPT-6.1 Sol and GPT-6 Astra respectively; neither incident reached the highest severity level P0.
The Decoder
OpenAI GPT-5.6 Sol and Google Gemini 4 Argon discover a critical PHP vulnerability affecting millions of websites
OpenAI's GPT-5.6 Sol and Google's Gemini 4 Argon discovered a critical PHP vulnerability affecting millions of websites in the Patch-the-Planet program, which was independently discovered by the security company Wiz using its own Wiz Atlas platform.
More worth reading
- The DecoderClaude Science designs protein binders from scratch to combat lethal scorpion venomAnthropic's Claude has essentially designed a protein complex targeting scorpion neurotoxins in the Claude Science research environment, a type of toxin that kills over 3250 people annually.
Hugging Face BlogOfficialNVIDIA releases Kumo Relational relational foundation model, capable of prediction without trainingNVIDIA releases pre-trained relational foundation model Kumo Relational, enabling direct prediction on enterprise databases without task-specific training, feature engineering, or table flattening. The model outperforms supervised relational deep learning and expert feature engineering gradient boosting trees on 41 tasks across 15 databases, achieving an MRR 8% higher than AutoGluon ensemble on the SAP SALT benchmark.
Tomasz TunguzSpaceX and DeepSeek's inference business revenue data reveals GPU price changes and market trends.SpaceX's third-quarter report and DeepSeek's inference business disclosure show that the inference business remains healthy under current prices and efficiency. DeepSeek's theoretical maximum revenue reaches $38 per card-hour, but the spot price of B200 is $8.77, while SpaceX's contract price is $5.27, achieving revenue of $2.44 per card-hour. AI companies are pushing Qualcomm to run billion-parameter models on smartphones, but data center construction lags, GPU inventory accumulates, and demand continues to drive AI revenue growth, which may offset the pressure of price declines.
TechCrunch · AIWalmart launches Agent Gateway, welcoming AI agents, in contrast to AmazonWalmart releases Agent Gateway to manage transactions, welcoming AI agents shopping, in contrast to Amazon's blockade of AI agents. The technology supports connection via MCP protocol, allowing AI agents to search Walmart inventory and make purchases on behalf, while also launching Agent Profiles control features, enabling users to authorize and restrict AI agent behavior. Walmart emphasizes that agents must comply with specific terms, including not exceeding user intent, inability to access payment information, and violators will be blocked.
NVIDIA Technical BlogNVIDIA Sentry: A Runtime Security AI Agent FrameworkThe NVIDIA Sentry framework prevents AI agents from bypassing security boundaries by monitoring them externally at runtime. The framework can monitor the tools, credentials, and services used by the agents, ensuring they do not compromise security controls while completing their objectives. NVIDIA Sentry, as a security software framework, is designed to protect AI agents, preventing them from tampering with or disabling monitoring mechanisms.
Modal BlogModal Dojo: A library to easily start production-level LLM training.Modal Dojo is a library for easily performing production-level LLM training, supporting models from 500 million to 10 trillion parameters, providing automatically optimized training recipes and built-in observability dashboards. Users can select models, load data, and start training through simple prompts, supporting data sources such as Hugging Face and Harbor environments, and can adjust recipes through parameters.
