Friday, Nov 13, 2026
AI;DR
2026-11-13 · 14 stories- googleGoogle launched Gemini 4 Hydrogen for rapid execution across diverse agentic workflows, priced at $1.50/1M input tokens and $7.50/1M output tokens. Artificial Analysis says it nearly matches Claude Sonnet 5.5 and GPT-6.1 Sol, with xhigh reasoning, at one-third Sonnet’s cost and about 2.5x Sol’s.+1
- deepseekForensic analysis found evidence that DeepSeek, Moonshot AI, MiniMax and Ant Group accessed export-controlled NVIDIA Blackwell GPUs. A Reuters review of more than 150 contracts found H100, H200 and H800 chips underpinning over 23 billion yuan ($3.2B) in Chinese AI projects, while IDC estimated NVIDIA held 34% of accelerator shipments in the first half of 2026.+1
- nvidiaJensen Huang warned finance executives that major cloud providers were shifting from NVIDIA to alternative chips while NVIDIA faced an “avalanche” of committed wafer and memory shipments. The turmoil preceded Colette Kress’s appointment as CEO and disputes over a $250B OpenAI credit plan cut to $120B and a failed $500B financing vehicle.
- llmEdge0-145B is an open-source 145B model for local coding, reasoning and agent tasks. It runs inference on a personal computer, eliminating cloud-model token fees for local use.
- agentExpo launched Expo Simulators, letting an AI agent run a cloud iOS simulator, install an app, perform interactions and produce recordings and screenshots before code reaches developers. Shared links let humans test the exact build, addressing the review bottleneck in agentic coding.
- nvidiaCommerce Secretary Howard Lutnick agreed to testify before the Senate Banking Committee on Dec. 10 about failures to keep advanced AI chips out of China; Jensen Huang was separately asked to appear on Dec. 3. The scrutiny follows claims that NVIDIA’s Tim Teter edited draft export policies and that Megaspeed bought more than $4.8B of NVIDIA hardware.
- inferenceSouth Korea-based DeepX taped out DX-M2, an edge AI chip that runs inference for 100B-parameter LLMs on robots and drones in under 5 W. The company targets mass production in fall 2027 for robotics and other edge markets.
- llmHugging Face released Transformers v5.23.0 with Inkling2 Small, a multimodal model from Thinking Machines Lab with 594B total and 17B active parameters, plus JEV, a decision head for scoring single-token labels. Inkling2 accepts text and images and generates text across languages and coding languages.
- metaggml-org released llama.cpp v1.0.0 as a stronger foundation for the local AI ecosystem. The release targets developers building and running AI locally.
- open-sourceFunAI Lab released training code and weights for open vision models based on FLUX.2 Klein 4B and 9B. The models scored 73.9 mIoU on Cityscapes versus 69.9 for Google DeepMind’s closed-source Vision Banana, using public data and 148 checkpoints, with the new legend-ensembling test-time scaling method.
- openaiOpenAI told lawmakers it supports a bipartisan congressional effort to block China’s access to AI chips and chipmaking tools. The position would align the company with tighter controls on China’s AI infrastructure.
- agentBrave Search API now lets developers and AI agents pay per query in USDC through the open x402 protocol. Real-time settlement works without an account or API key, simplifying on-demand search access.
- anthropicNscale told investors it needs additional funding for 2027 commitments after ordering NVIDIA chips for a $45B Anthropic deal. The financing need casts doubt on its planned $35B IPO.
- nvidiaSenate Majority Leader John Thune said Congress has a “high probability” of imposing new China chip-export limits, citing support from Hakeem Jeffries, Chuck Schumer and Mike Johnson. Lawmakers may attach them to the National Defense Authorization Act or other must-pass legislation during the lame-duck session.