Alibaba launches CosyVoice Studio, a unified voice platform
The Qwen-Audio-based product bundles dictation, voice agents and audiobook generation, and ships free on iOS, Android, Mac and Windows.
August 6, 2026
All briefings and quick hits published that day.
- ⚡ Quick hits
- Industry A leaked new Microsoft Copilot icon shows the assistant moving to a flatter, simpler design. IT之家 ↗
- Industry South Korea's Mobilint launched the MLD-R1, a USB-form AI accelerator delivering 10 TOPS of INT8 compute for edge inference. IT之家 ↗
- Agents Cloudflare launched temporary accounts so agents can deploy Workers instantly via wrangler deploy --temporary, with deployments expiring in 60 minutes unless a human claims them. InfoQ 中文 ↗
- ⚡ Quick hits
- Open Source MiniMax said the community built a distillation LoRA for its open-weight H3 video model four days after release, cutting sampling from 20 steps to 4–8. X @MiniMax_AI ↗
- Industry Apple's bug bounty review team went offline under a flood of AI-generated reports, imposing submission caps and a 30-day cooldown from August 2. 量子位 ↗
Huawei's openJiuwen ships agent swarm into bank production
The open-source agent platform released an enterprise-grade distributed swarm architecture, first deployed in production at China's Postal Savings Bank.
Seedance 2.5 API opens as Western platforms line up to carry it
ByteDance switched on public API access for its 30-second video model; Luma had it live a day earlier and Runway shipped it on the day, with Krea, Lovart, Pika and others announcing integrations still rolling out.
- ⚡ Quick hits
- Agents Microsoft open-sourced code-testing-generator, a polyglot unit-test agent that completed 92.1% of 152 internal tasks versus 78.9% for stock GitHub Copilot. MarkTechPost ↗
- AIGC Hailuo AI demoed redubbing existing clips into another language or tone while preserving the original speaker's exact voice and natural speech cadence. X @Hailuo_AI ↗
- Models Liquid AI released LFM2.5-2.6B, an on-device agentic model with 128K context, tool calling and open weights in native, GGUF, MLX and ONNX formats. MarkTechPost ↗
- Research Together AI benchmarked DeepSeek-V4 Flash-0731 against GPT-5.6 Luna on DeepSWE software engineering tasks, finding it delivers 80% of Luna's performance far more cheaply. X @togethercompute ↗
Anthropic loosens Claude Fable 5's biology safeguards
A rewritten classifier constitution cut biology-related fallbacks by roughly 85%, after months of complaints that the model refused routine health questions.
- ⚡ Quick hits
- Open Source Ant Group open-sourced Avernet under Apache 2.0, a multi-agent collaboration layer already deployed across 12 internal business units with over 90% task completion. 量子位 ↗
- AIGC MiniMax will livestream with ComfyUI on August 7 to stress-test H3, showing open weights, native stereo audio and clips up to 2K resolution and 15 seconds. X @MiniMax_AI ↗
- AIGC Alibaba opened public beta of video model Wan3.0, which generates 30-second clips and accepts docs, spreadsheets, slides and PDFs as input. X @Alibaba_Wan ↗
- Industry Genspark argued that enterprise AI bills keep climbing even as per-unit intelligence got roughly ten times cheaper this year, because agents moved into production. X @genspark_ai ↗
- Industry Replit said a live session where 14,075 people built apps together set a Guinness World Record for the largest AI video lesson. X @Replit ↗
- Industry OpenAI's first Jony Ive-designed device is reportedly a hockey-puck-sized smart speaker priced between $300 and $400, according to Bloomberg's Mark Gurman. TechCrunch ↗
- Agents OpenAI put Codex Security Review into research preview, having the agent inspect GitHub pull requests for security issues using full repository context. X @OpenAIDevs ↗
AMD buys Taalas to etch AI models directly into silicon
The Toronto startup builds chips whose metal layers are finalised around one model's weights, promising inference without HBM, advanced packaging or liquid cooling.
- ⚡ Quick hits
- Agents Perplexity made GPT-5.6 Terra the default model for all Perplexity Computer subagents, saying it scores eleven points above Sonnet on WANDR. X @perplexity_ai ↗
- Models Moonshot AI's open-weight Kimi K3 became generally available in GitHub Copilot for agentic coding, hosted by Fireworks AI. X @github ↗
OpenAI splits ChatGPT tiers between GPT-5.6 Sol and Luna
Paid ChatGPT users get an updated GPT-5.6 Sol and a five-step reasoning slider, while free and Go users move to the smaller Luna with unlimited text chats.
- ⚡ Quick hits
- Funding Naïve raised $28.5 million to automate the administrative grunt work of incorporating and running a company. TechCrunch ↗
OpenAI, Vercel and rivals ship Agent Plugins 1.0 standard
A vendor-neutral format bundles Agent Skills and MCP server configs into one portable plugin, supported at launch by ChatGPT, Codex, Cursor, Copilot, Kiro and VS Code — with Google joining the steering committee the same day.
- ⚡ Quick hits
- AIGC Meshy upgraded Multi-View Image to 3D so textures are now generated from multiple reference views rather than a single one. X @MeshyAI ↗
- Agents Cursor detailed its Router, which classifies requests and routes them across Grok 4.5, GPT-5.6 Sol and Opus 5 to cut latency and cost. X @cursor_ai ↗
- AIGC ChatGPT can now call Adobe Photoshop and Lightroom directly, letting users retouch and design inside the chat without leaving the conversation. IT之家 ↗
- AIGC ElevenLabs shipped Dubbing v2 on its API, adding regional accent accuracy and letting enterprises bring their own transcripts and regenerate only edited segments. X @ElevenLabs ↗
DeepMind open-sources WeatherNext cyclone forecasting model
A Nature paper puts Google DeepMind's WeatherNext Cyclones ahead of operational systems by over a day of lead time, with code and weights released on GitHub.
Cloudflare's Kitesurf drops Chromium for agent browsing
The agent-first browser runs in V8 isolates on Workers with no Chromium underneath, claiming 3–7x lower CPU and memory on common agent tasks at 1.7–1.8x the wall-clock time.
- ⚡ Quick hits
- Open Source Unsloth released DSpark, which makes DeepSeek-V4-Flash-0731 GGUF builds generate roughly 1.4 to 2 times faster locally with no accuracy change. X @UnslothAI ↗
- Research A community effort pointed coding agents at ICML 2026, with over 1,200 people reproducing more than 2,000 papers in the largest conference reproducibility audit yet. X @Gradio ↗
Suno to watermark and fingerprint AI-generated songs
The music generator will embed tamper-resistant audio watermarks, add fingerprinting, restrict bulk downloads and license Musixmatch's Sentinel for lyric copyright detection.
- ⚡ Quick hits
- AIGC Luma Labs added MiniMax H3 to Luma Agents, generating up to 15 seconds of 2K video with native stereo sound. X @LumaLabsAI ↗
- Open Source Shanghai AI Lab open-sourced Mobius, an architecture it frames as upgrading knowledge connectivity the way Transformers upgraded token connectivity over RNNs. X @intern_lm ↗
- Open Source Genspark brought its open-source GenOffice suite to Linux, matching the existing Mac and Windows builds. X @genspark_ai ↗