Nvidia Reportedly Nears $12.9B Hugging Face Deal
Conflicting reports say Nvidia has agreed, or is close, to acquire the central hub of the open-model ecosystem for $12.9 billion.
August 26, 2026
All briefings and quick hits published that day.
- ⚡ Quick hits
- Industry MiniMax reported annual recurring revenue above $800 million, with enterprise customers contributing 80% of revenue. InfoQ 中文 ↗
- Agents Harper recreated the same AI-built application deployed on Vercel and reported performance differences reaching 14-fold in its tests. InfoQ 中文 ↗
- AIGC Meshy previewed Meshy 7, claiming stronger alignment between uploaded source models and generated assets intended for 3D printing. Meshy on X ↗
- Agents Wan published a builder guide explaining how creators can turn Wan-powered generation workflows into reusable agent skills. Alibaba Wan on X ↗
- AIGC Alibaba launched a weekly Wan Skill Challenge rewarding three creators who package Wan-powered video workflows as reusable skills. Alibaba Wan on X ↗
- Agents MongoDB released its managed Atlas MCP server as a Claude Code plugin, enabling agents to query data and optimize vector search. MongoDB on X ↗
- Models Qwen Cloud priced Qwen3.8-Flash at $0.15 per million input tokens, $0.47 for output and $0.016 for cache hits. Qwen on X ↗
- ⚡ Quick hits
- Models Qwen3.8-Flash became available through OpenRouter for coding assistants, agent workflows and long-video understanding. Qwen on X ↗
- Funding Chinese inference-platform startup Key Motion raised tens of millions of yuan and introduced a model-routing service positioned as a domestic OpenRouter. 量子位 ↗
- Open Source Qwen received day-zero TokenSpeed support covering its new GDN, QSA, N-gram embedding and FP8 architecture components. Qwen on X ↗
- Industry OpenAI launched commercial operations in Brazil, establishing a São Paulo team to work with businesses, developers, researchers and public institutions. OpenAI ↗
- Agents Siemens detailed an industrial AI agent that embeds the company’s engineering knowledge rather than merely wrapping a general-purpose language model. 量子位 ↗
- Safety Bill Gates warned that advanced AI could threaten humanity if governance and technical safeguards fail to keep pace with capabilities. SiliconANGLE ↗
- Funding AI startup Instinct raised $350 million at a $2.5 billion valuation following rapid viral adoption of its consumer product. TechCrunch ↗
AWS Orders Two Million More Nvidia GPUs for 2027–28
Amazon will deploy two million additional Blackwell Ultra and Rubin-generation GPUs while extending Nvidia technology across AWS and robotics.
Claude Cowork Adds an Isolated Built-In Browser
Anthropic is giving its desktop work agent a separate browser that can navigate sites and complete forms without controlling personal tabs.
- ⚡ Quick hits
- Funding Deep Cogito raised $43 million to develop AI models designed to improve themselves through iterative learning. SiliconANGLE ↗
Anthropic Secures $45B in Nscale Compute Capacity
The six-year agreement ties Anthropic’s next growth phase to 460 megawatts of Nvidia Vera Rubin infrastructure in West Virginia.
- ⚡ Quick hits
- Open Source Hugging Face began using the open-weight Inkling-Small model to turn research-paper abstracts into concise summaries. X (@huggingface) ↗
- Agents GitHub introduced Agentic Workflows for automating code changes, organization-wide rollouts and code-quality assessments at scale. X (@GitHubNext) ↗
- Agents AWS added AgentCore Evaluations for assessing agents built with any framework through configurable evaluators and production monitoring. AWS Machine Learning Blog ↗
- Research Google Research introduced GlucoFM, a foundation model for extracting health insights from continuous glucose-monitoring data. Google Research ↗
- Agents VS Code added a rubber-duck command that asks a complementary model to review agent work for missed details and edge cases. X (@code) ↗
Recraft V4 Styles Adds Reference-Led Generation
The design model can derive a reusable visual style from one to ten reference images without requiring a separate training process.
- ⚡ Quick hits
- AIGC Runway added Meta’s Muse Image to its platform alongside its existing image and video generation models. X (@runwayml) ↗
- Agents Bolt introduced an agentic security audit that scans applications, applies patches and hardens detected vulnerabilities. X (@boltdotnew) ↗
- Agents Gemini Live added autonomous multi-step tasks, Gmail actions, Daily Brief access and personalized assistance based on previous conversations. X (@GeminiApp) ↗
- Research Anthropic gave independent researchers access to privacy-preserved Claude usage data for studying collaboration, wellbeing and real-world productivity. X (@AnthropicAI) ↗
- Models Harvey and Fireworks post-trained Tenet from Kimi K3 for long-horizon legal work, nearly doubling fully passed benchmark tasks at similar cost. X (@FireworksAI_HQ) ↗
- Research Goodfire presented a more token-efficient method for identifying chain-of-thought branching tokens that materially alter model outcomes. X (@GoodfireAI) ↗
Google Launches Gemini 3.5 Transcribe in Preview
Google’s new speech model handles live multilingual audio, cleans disfluencies and powers voice-driven workflows in products such as Gemini for macOS.
- ⚡ Quick hits
- AIGC Harvard Business School used HeyGen LiveAvatar to create interactive AI avatars of professors for its Foundry entrepreneurship program. X (@HeyGen) ↗
- Agents Devin sessions can now launch managed subagents with separate virtual machines, and those child sessions can spawn further subagents. X (@cognition) ↗
Qwen Opens Flash-Next as a Preview of Qwen4
Alibaba’s 125B-parameter multimodal model activates 6B parameters per token and introduces the architecture intended for Qwen4.
Z.ai Releases GLM-5.3-Flash With Open Weights
The 320B mixture-of-experts model pairs native multimodality and one-million-token context with aggressively priced hosted inference.