Tencent Hunyuan ships Hy ASR 3.0 preview via cloud API
Tencent's new speech model reports about 3% word error rates in Mandarin, English and Cantonese, and is live on Tencent Cloud and in Yuanbao.
August 3, 2026
All briefings and quick hits published that day.
- ⚡ Quick hits
- Industry OpenAI countered Apple's trade-secret suit with iMessage logs showing Apple staff repeatedly texted a departed engineer for internal information, arguing Apple's own access controls failed. The Decoder ↗
- ⚡ Quick hits
- Open Source Reflex open-sourced XY under Apache-2.0, a Rust-backed Python charting library holding ~0.08s renders from 10,000 to 100 million interactive points. MarkTechPost ↗
- AIGC Recraft showed V4.1 swapping an object's material while keeping its form intact, pitching editable physical logic rather than a freshly regenerated image. X @recraftai ↗
- Research A Kansas mathematician refuted OpenAI's claimed disproof of the Connes rigidity conjecture within 24 hours, tracing 37,000 lines of Lean 4 to a group failing preconditions. 量子位 ↗
- Industry OpenAI's luxury Hudson Valley creator retreat drew backlash over $2,000-a-night rooms and farm activities, with several influencers deleting their posts afterwards. 量子位 ↗
- Agents MarkTechPost detailed a skill-auditing pipeline using NVIDIA SkillSpector and LangGraph to scan agent skills for shell injection, credential harvesting and MCP attacks before deployment. MarkTechPost ↗
- Funding Om AI raised several hundred million yuan led by Qianhai Mother Fund for edge-native multimodal perception models, whose Homer AI serves nearly 100,000 visually impaired users. 量子位 ↗
- Agents Y Combinator open-sourced QM under MIT, a multiplayer agent harness for Slack and web giving each person isolated memory, files, permissions and sandboxes. MarkTechPost ↗
- Policy MiniMax rebutted claims its H3 video model cannot be legally used in certain regions, saying it can be licensed for deployment in the US, EU, UK and South Korea. X @MiniMax_AI ↗
- Safety A study of 414 internet-facing MCP servers found 68 reportable flaws and 91.8% lacking OAuth, plus 687 tools offering unrestricted shell execution. arXiv ↗
Alibaba's Wan adds real-time Camera-to-Image generation
Wan launched a Real-time feature suite whose first capability turns a live phone camera feed into stylised imagery, starting with anime looks.
- ⚡ Quick hits
- AIGC PixVerse showed a rooftop-to-office scene built in PixVerse Canvas, saying the physics holds up down to a tote bag swinging correctly through the shot. X @PixVerse_ ↗
- Industry Sakana AI became a full member of Japan's AI Robot Association this month, collaborating on world models and physical AI research under its RSI Lab. X @SakanaAILabs ↗
- Models Together AI put DeepSeek V4 Flash into production, citing stronger terminal, repository and full-stack coding plus better tool use for long-running agents. X @togethercompute ↗
- Industry Palantir CEO Alex Karp called AI labs' enterprise ambitions "Marxist" after Q2 revenue hit $1.9 billion, up 93%, with $1.1 billion profit. TechCrunch ↗
MiniMax H3 tops Video Arena as No.1 open video model
Video Arena placed MiniMax-H3 first among open models in both text-to-video and image-to-video, some 280 Elo points clear of the next open weights entry. The weights are now public, but the licence restricts open-weight use in the US, EU, UK and South Korea.
- ⚡ Quick hits
- Research MarkTechPost published a hands-on guide to evaluating multimodal vision models with Moonshot's PerceptionBench, using robust data loading and automated LLM judging. MarkTechPost ↗
- Industry A US company's AI now lets Ukraine's low-cost kamikaze drones track targets autonomously, a step toward future swarm attacks. Ars Technica ↗
- Industry OpenAI published an engineering deep dive on rebuilding its voice stack for GPT-Live, routing audio down a fast path and cutting session startup from six network round trips. X @OpenAI ↗
Cursor agents gain read-write access to Google Workspace
Cursor shipped plugins letting its agents read and write across Gmail, Drive, Calendar, Docs, Sheets and Chat, pushing a coding tool into general knowledge work.
Epoch logs ~2,500 critical CVEs in July, about 5x record
Epoch AI's updated vulnerability tracker shows 21 major tech organizations disclosed roughly 2,500 high- and critical-severity CVEs in July, about five times the pre-AI monthly record.
- ⚡ Quick hits
- Agents Vercel launched the v0 API, giving developers programmatic app building: start a chat from a prompt, repo or ZIP, and render a dev-server preview. X @v0 ↗
- Funding The team behind Design Arena raised a $7.9 million seed led by Index Ventures to rank media-generating AI models on human taste judgments. TechCrunch ↗
- Policy An AI-supervised remote exam failed so badly that 58,000 students have been ordered to retake it, renewing scrutiny of automated proctoring. Ars Technica ↗
- AIGC Meshy rolled out 8K texturing for its 3D generation platform, showcasing worn metal, plush and plastic materials rendered at production fidelity. X @MeshyAI ↗
- Open Source Xiaoda Robotics and NTU's S-Lab open-sourced ACE-Data-0, an embodied dataset of 17 million frames across 200 task categories recorded in two real homes. InfoQ 中文 ↗
- Industry Cognition hired former Coinbase chief legal officer Paul Grewal, who stepped down on July 31, as its chief legal and global affairs officer. X @cognition ↗
- Open Source Microsoft Research open-sourced Orchard, a Kubernetes-based framework for training and evaluating agents, with its 3B-active-parameter Orchard-SWE hitting 69.7% on SWE-bench Verified. Microsoft Research ↗
- Safety Interpol said AI has become the core operational driver of cybercrime across Africa, powering scams, fraud and impersonation at growing scale. The Decoder ↗
- Agents ElevenLabs launched ElevenAgents for enterprises, saying over 10 million voice and chat agent conversations already run on the platform weekly with configurable guardrails and regional data residency. X @ElevenLabs ↗
- AIGC Kuaishou's Kling published a Kling MCP tutorial showing agents batch-producing food promo videos, from analyzing creative styles and building storyboards to generating final shots. X @Kling_ai ↗
- AIGC SenseTime open-sourced SenseNova U1.5-Lite-Preview, an 8B unified multimodal model that generates native 4K images and supports pixel-level editing, released on GitHub, Hugging Face and ModelScope. 量子位 ↗