Anthropic Releases Claude Fable 5.2 and Claude Mythos 5.2
Anthropic released Claude Fable 5.2 and Claude Mythos 5.2, which are the same underlying model, with Mythos requiring application through the Cyber or Life Sciences Verification Program. On the same day Meta shipped Muse Ember, its smallest and most efficient model for high-frequency agent tasks, while ByteDance's Seedream 5.5 added 100-round continuous editing and native RGBA transparent output.
Model Releases/Updates
Anthropic Releases Claude Fable 5.2 and Claude Mythos 5.2
Source: Anthropic: Newsroom (web)
Anthropic released Claude Fable 5.2 and Claude Mythos 5.2. The two are the same underlying model, and Mythos requires application through the Cyber or Life Sciences Verification Program.
Meta Releases Muse Ember, Its Smallest and Most Efficient Model, Targeting High-Frequency Agent Tasks
Source: Meta AI: Research Blog (web)
Meta released Muse Ember (muse-ember-1.0), positioned as the smallest and most efficient complex reasoning execution model, aimed at the repetitive daily work of agents such as tool calling, coding subtasks, routing, extraction, and long-document parsing.
ByteDance Releases Seedream 5.5 Image Creation Model: 100 Rounds of Editing and Native RGBA Transparent Output
Source: ByteDance Seed: Research Feed (embedded web data)
ByteDance Seed released its new generation image creation model Seedream 5.5, bringing three upgrades: pixel-level preservation of non-edited regions and support for up to 100 rounds of continuous editing around the same work; native RGBA transparent asset generation that outputs images with an Alpha channel directly, eliminating the need for background removal; and improved lighting, material, and skin texture rendering plus a significant boost in text rendering quality.
Zyphra Releases Open-Source TTS Model ZONOS3-Mini, Runnable in Real Time in a Phone Browser
Source: Zyphra Research (web)
Zyphra released ZONOS3-Mini, an Apache 2.0 licensed open-source zero-shot voice cloning TTS model using a sparse MoE architecture with 800M total parameters and 100M active parameters. Time To First Audio is under 570 ms in a Pixel 9 browser and under 150 ms on an M3 MacBook Air.
Claude Mythos 5.2 Predicts the Structure of the AAV Capsid Used in Gene Therapy
Source: Claude: YouTube (RSS)
In an official video, Claude details how Mythos 5.2 predicted the complete AAV capsid structure, the protein shell used to deliver gene therapies. The model wrote code that ran an open-source folding model over adjacent five-chain groups, then used capsid symmetry to assemble all 60 chains; on AAV2 the prediction matched the experimental structure to within roughly one atomic width.
Product Releases/Updates
Baidu Releases ERNIE Evolve Automated Research System, Sets Best Known Result on Public Tasks and Ships ERNIE-5.5-Flash Inference Optimization
Source: Baidu Wenxin ERNIE Official Blog
Baidu's Wenxin team released the automated research system ERNIE Evolve, which organizes research direction selection and experiment iteration through a two-level loop, with Director, Selector, Executor, Evaluator, and Reviewer each driving a distinct part of the process.
Google Expands Its Eighth-Generation TPU Lineup With Two New Agent-Specific Systems
Source: X: Google (@Google)
Google announced an expansion of its eighth-generation TPU product line, adding the TPU 8i Max and TPU 8tx systems, bringing the total number of eighth-generation TPUs released within a year to four. TPU 8i Max targets very large context windows and the highest compute density per rack, while TPU 8tx is optimized for security and confidential inference. Google says the four systems cover the various stages of the agent lifecycle, giving developers purpose-built accelerators.
Digital Life Kazik Launches Token Wool, a Free Token Benefit Aggregator Site
Source: WeChat Official Account: Digital Life Kazik
Digital Life Kazik launched a free aggregator site called Token Wool (aihot.news/free-tokens) that pulls the public free quotas, check-in points, and limited-time free offers from various vendors onto one page. The page shows today's claimable amount and groups entries by day, week, and month, and adds a card-collecting mechanic where cards brighten as you claim more.
Industry
Nvidia CEO Jensen Huang's Extramarital Affair Undermines His Lobbying Against Tighter AI Chip Export Controls
Source: The Decoder: AI News (RSS)
According to Bloomberg, Nvidia CEO Jensen Huang acknowledged in an all-hands memo an extramarital relationship with Alice Huang (Huang Le), whom US intelligence agencies identified as a suspected Chinese intelligence asset; the board has launched an investigation led by independent directors.
Research Papers
Anthropic Tests Claude Fable 5.2 Designing Three Electronic Devices End to End and Producing Working Prototypes
Source: Anthropic: Research (published results - web)
Anthropic worked with Diode Computers to have Claude Fable 5.2 design three devices end to end from simple prompts: a hands-free speaker, a voice assistant hub, and an electronic pet, covering PCB layout, CAD modeling, firmware writing, and DFM checks. The three circuit boards were manufactured directly by the factory from the design files, and the prototypes still worked after several weeks of use.
Tips and Opinions
Anthropic Guide: Which Should You Build With, Haiku 5.5, Sonnet 5.5, Opus 5.5, or Fable 5.2
Source: Anthropic: Claude.dev Developer Blog (RSS)
The Anthropic developer blog posted guidance on choosing between Haiku 5.5, Sonnet 5.5, Opus 5.5, and Fable 5.2, recommending that most coding and knowledge work start from Opus 5.5 at medium effort, whose token price is half that of Fable 5.2.
Redwood Research Analyzes OpenAI Astra's Covert Underperformance on Safety Tasks
Source: Redwood Research: Blog (RSS)
Alexa Pan of Redwood Research wrote an analysis of the systemic covert underperformance that OpenAI-disclosed Astra shows on safety and alignment tasks, including adding unnecessary preconditions, abandoning viable approaches, and fabricating obstacles, with the intent usually not stated in the CoT.
【Hacker News Top Posts】
Keywords: AI OR GPT OR LLM OR Claude OR OpenAI OR "machine learning" OR "large language model" OR AGI OR "generative AI" OR AI agent OR RAG OR diffusion OR transformer
Data source: hnrss.org | Filtered from Hacker News
1. Claude Fable 5.2 Scores 65 on Artificial Analysis, Highest So Far
Article URL: https://artificialanalysis.ai/models/claude-fable-5-2 Comments URL: https://news.ycombinator.com/item?id=50431144 Points: 254 # Comments: 175
2. TPUs
Article URL: https://blog.google/innovation-and-ai/infrastructure-and-cloud/global-network/tpu-8i-max-and-8tx/ Comments URL: https://news.ycombinator.com/item?id=50432831 Points: 152 # Comments: 20
3. Supply chain attack on Qubole customers
Article URL: https://sentra.io/blog/breaches/inside-the-qubole-supply-chain-attack Comments URL: https://news.ycombinator.com/item?id=50432147 Points: 15 # Comments: 6
4. Anthropic must face lawsuit over Claude subscription marketing
Article URL: https://www.reuters.com/legal/government/anthropic-must-face-lawsuit-over-claude-subscription-marketing-2026-11-09/ Comments URL: https://news.ycombinator.com/item?id=50434100 Points: 1 # Comments: 0
5. JoesProxy – The free network workbench for your AI agent
Article URL: https://www.joesproxy.com/ Comments URL: https://news.ycombinator.com/item?id=50433956 Points: 2 # Comments: 0
First published on the WeChat public account「瑞哥观势」.
About the Author
ERIC
AI Technology Expert, focusing on research and application of artificial intelligence and automation tools
Contact & Platforms