Morning Brief: Volcano Engine releases Seedream 5.5, Apple uses simulation to train humanoid robots
Volcano Engine released the Doubao image creation model Seedream 5.5, which supports image generation, editing, cutout, and native RGBA transparent asset generation via natural language. Editing a single 2K square image costs 0.28 yuan, 60% cheaper than Seedream 5.0 Pro and nearly 40% faster, with the API launching on Volcano Ark this week. Apple's research team published a paper on arXiv introducing HALO, a humanoid robot control system, tested on a modified Unitree G1, achieving an average success rate of 91% across 360 tests in 6 unseen real-world scenarios. Anthropic released Claude Fable 5.2 and Mythos 5.2 on the same day, with the standard API priced at $8/$40 per million input/output tokens, a 20% price cut compared to 5.1.
Body · AI translation
Tuesday 10 Morning Brief: Volcano Engine releases Seedream 5.5, Apple research team trains humanoid robot with simulation, and more. Do you all really have to release new models on the same day? 8-minute read Morning Brief at a Glance
Volcano Engine releases Seedream 5.50
Apple research team trains humanoid robot with simulation1
Anthropic releases Claude Fable 5.2 and Mythos 5.21
Automattic launches AI app hosting platform Spacefast1
On November 9, Volcano Engine released the Doubao image creation model Seedream 5.5. On a pixel-level controllable basis, the model can use natural language to complete professional image generation, editing, and cutout. Seedream 5.5 can maintain consistency in character, product, scene details, and asset proportions even after multiple rounds of image editing. The model supports native RGBA transparent asset generation. Compared with Seedream 5.0 Pro, Seedream 5.5 can generate more realistic, natural, and cinematic photorealistic images. In terms of pricing, Seedream 5.5 uses two-stage billing of "input image + output image"; text tokens are free, and editing a single 2K square image costs 0.28 yuan, which is 60% cheaper than Seedream 5.0 Pro and nearly 40% faster to generate. The Seedream 5.5 API will launch on Volcano Ark this week. Source3
Apple research team trains humanoid robot with simulation
On November 4, the Apple research team submitted the paper Perceptive Humanoid Motion from Simulation to arXiv, introducing the humanoid robot control system HALO. The researchers tested this system on a modified Unitree G1. The robot can combine natural language instructions and camera images to walk to a specified object, turn sideways to pass through a half-open door, or crawl under a low obstacle. According to the paper, the team synthesized about 300 hours of primary training data in 1,000 procedurally generated indoor scenes. The team conducted 360 single-task tests in 6 real scenes that were unseen during training and had not been pre-mapped, and the robot's average success rate was 91%. The team also tested 3 combinations of consecutive actions in a real scene, with 10 tests each, and the average success rate was 70%. Source
HALO? Does Microsoft have no objection? 12
Anthropic releases Claude Fable 5.2 and Mythos 5.2
On November 9, Anthropic released Claude Fable 5.2 and Claude Mythos 5.2, upgrading programming, reasoning, and visual understanding capabilities. The two share the same underlying model. Fable 5.2 has restrictions on high-risk cybersecurity and biology tasks and has been launched on channels including the Claude app and API; institutions with related research needs can apply to use Mythos 5.2 through Anthropic's review program. The standard API price for both models is $8 per million input tokens and $40 per million output tokens, a 20% reduction from version 5.1. Anthropic says that output speed in standard mode is up to 30% faster than the previous generation, while the optional fast mode can further increase output speed to up to 2.5 times that of standard mode, at twice the unit price of standard mode. In the various standard tests published by Anthropic, Fable 5.2 and Mythos 5.2 scored higher than the 5.1 model, with improvements ranging from 4% to 38%; in another early experiment, about 1,000 agents running an early version of Fable 5.2, after being given intermediate propositions and some already-proven lemmas, completed a proof of Fermat's Last Theorem in about 30 hours. Source Has Fermat's theorem become a unit of measurement?
Automattic launches AI app hosting platform Spacefast
On November 9, Automattic, the parent company of WordPress, launched Spacefast, providing deployment and hosting services for websites and apps written by AI under natural language direction. Users can continue to use tools such as ChatGPT, Claude, or Cursor to deploy generated projects directly to Spacefast, or connect an existing Git code repository. Spacefast runs on WordPress infrastructure and supports custom domains and a built-in database. Deployed projects are private by default; users can invite team members via link to view them and leave comments on the page, and then have AI modify them based on the feedback. The platform offers a free plan, with paid plans starting at $5 per month. Source No glasses: Anthropic; Glasses: Automattic
Tencent Yuanqi stops service
Tencent Yuanqi Platform officially ceased service on November 9; all public-facing services, including app access, data reading and writing, API calls, and console use, no longer support user access. Agents published to other channels via the Yuanqi platform (including Official Accounts, Service Accounts, WeChat Customer Service, etc.) no longer support responding to user conversations. Since the 29th of last month, Tencent Yuanqi has stopped creating new agents and distributing API Keys. Users can currently submit export task requests through the data export console. After January 9, 2027, data export requests will no longer be accepted, and except for logs and other records that must be retained as required by laws and regulations, the official side will delete all user data. Tencent Yuanqi is a zero-code agent development and distribution platform launched by Tencent. Based on the Hunyuan model, users can quickly develop dedicated AI chat agents that can reply automatically. Source Never heard of it...
Microsoft removes PC-to-PC migration feature
Previously, Windows 11 offered a PC-to-PC migration feature that allowed users to transfer files and settings directly between two PCs over Wi-Fi or Ethernet. The tool has now been found to be removed. Microsoft's updated support documentation recommends that users use the Windows Backup tool. This tool no longer relies on a direct connection between two PCs, but is based on the cloud storage service OneDrive. Windows Backup requires users to sign in to a personal Microsoft account and back up the supported folders Desktop, Documents, Pictures, Videos, and Music to OneDrive. Windows Backup also records Wi-Fi networks, passwords, and some system settings, which helps speed up configuration on a new device. The app list is also recorded, but apps still need to be reinstalled. Source Optimized the problem of the original migration being too smooth. 1349
Instagram age estimation feature rolls out globally
On November 9, Meta announced the global rollout of Instagram's AI age estimation feature. Even if the birthday a user enters indicates they are an adult, as long as the system determines they may be a teenager, teen protection settings will automatically be enabled for the account. Meta expects this will bring millions of additional teen users under protection. The feature will also roll out gradually on Facebook, and a similar feature will launch on Threads later this year. Instagram teen accounts restrict by default who can send direct messages to the user, reduce recommendations of content unsuitable for their age, and provide usage time reminders. Users under 16 need parental permission to relax the default protection settings. Source What about adults who are still young at heart? 5
Cyberspace Administration of China launches unified filing system
On November 9, the Cyberspace Administration of China launched the Internet New Technology and New Application Information Service Filing System. Enterprises can file matters such as generative artificial intelligence, algorithms, deep synthesis, and blockchain on the same platform, and the public can query filing information online. The original Internet Information Service Algorithm Filing System and the Blockchain Information Service Filing Management System were discontinued at the same time. Source
Checking Apple's filing is indeed more convenient now 1
Adobe and Disney reach strategic cooperation on AI tools
On November 9, Adobe and The Walt Disney Company announced a multi-year cooperation agreement at Adobe MAX. The two sides will jointly develop generative AI models through Adobe Firefly Foundry, customizing models with materials such as character settings, scenes, and visual styles from Disney's works, for use in early-stage creative design for theme parks, cruise line projects, and film and television studios. Adobe will also give Disney early access to the latest features of tools such as Firefly, Premiere, and Photoshop, and provide engineering and research support. Disney Imagineering's R&D team has already used Firefly Foundry for theme park design. Cooperation projects between Adobe and Walt Disney Studios' innovation lab StudioLAB and ESPN Edge have also launched. StudioLAB will explore new internal creative workflows with Adobe, while ESPN Edge will study how to use AI to improve the presentation of sports content and make ESPN's historical program archives more accessible to viewers. Source
Meta releases Muse Ember model
On November 9, Meta released Muse Ember, its smallest multimodal model to date, now available on Muse Code and the Meta Model API. The model was trained using Muse Wonder Full and is mainly used to perform tasks that AI agents need to handle repeatedly, including calling tools, executing programming subtasks, extracting information, and parsing long documents. Muse Ember supports text, image, audio, video, and PDF input, with a context window of about 1 million tokens and a maximum output of about 520,000 tokens per request. It charges $0.10 and $0.50 per million input and output tokens, respectively. According to data published by Meta, Muse Ember has a composite score of 35.8, close to Claude Haiku 5.5's 36.2 and higher than GPT-6 Luna's 28. Source 2
Brief news worth a look
Perplexity launched the Teammates agent for Perplexity Computer, allowing users to hand over work that requires long-term follow-up, such as tracking competitor developments and planning and coordinating projects. Teammates can arrange steps on its own, process tasks in parallel, and then hand the results to the user for review. Perplexity Max individual subscribers who have obtained research preview access can already use the feature on the web, and an iOS version will be launched later. Source
On November 6, Meta began rolling out dedicated models to Muse and Meta AI users. This update uses Muse Wonder to train existing models so that they can retrieve key information from past conversations when the current task requires it, and try other methods when encountering obstacles in executing tasks. Meta's proactive memory evaluation shows that Muse's new model correctly used key context in 90% of test tasks, compared with only 76% for the old model. Source
Samsung may launch a paid option for the Galaxy AI features built into its phones. The Galaxy AI Premium subscription may have three tiers: Starter, Plus, and Pro, corresponding to 10,000, 20,000, and 50,000 AI points, respectively. The free tier will be limited to 10 translations, 5 text summaries, and 2 image generations per day. Existing features will remain free, while new features such as Bixby Action and video generation will be available only to users on the highest paid tier. Source
Apple has begun offering an in-app purchase for an "Intelligent Usage Pack" in the latest version of the Apple Creator Studio app, priced at $49.99 and valid for 12 months. Source
On November 9, Wangxin Shanghai issued an announcement on changes to the filed information of generative artificial intelligence services. Apple Intelligence has added two new scenarios, Apple tablets and Apple laptops, and has completed filing. Source
Google Wallet for Android has added the overseas QR code payment feature Traveler QR. The feature currently supports only Singapore, where travelers can pay at offline merchants connected to its partner network through Alipay+. Source
Recent updates from Shaoshupai
We have opened a special page to display the awards and results of the "Outstanding Author Selection." The "Outstanding Author Selection" is a mechanism implemented by the editorial team since 2023. After the end of each quarter, it selects existing authors who made outstanding contributions during that quarter, as well as promising new authors who have emerged. Winners receive an exclusive badge and prizes. Go to view
Double 11 picks: Sonos Play 15% off with a charging kit, CD player with a free transparent cover. Shop now
Articles you may have missed
Double 11 picks: Sonos Play 15% off with a charging kit, CD player with a free transparent cover
How does the white box on the ceiling protect our safety?
A family's living room, and one person's late night — Sonos Beam Ultra and Ace Ultra review
Double 11 | Limited-time offers on Shaoshupai membership, genuine software, and paid columns
App Review | Apps worth paying attention to recently
> Download the Shaoshupai 2.0 client and follow the Shaoshupai Official Account to unlock a new reading experience 📰
> Practical, useful genuine software, presented to you by Shaoshupai 🚀 Pai Morning Post This article's editor:
@Beixiao
14 Pai friends have already charged