AI Signals Library for Builders

Definitions, verification, evaluation — and latest briefs

Start here (standard answers)

Explore the evergreen library

  • Topics — evergreen pages with evidence timelines
  • AI Answers — direct, quotable answers + FAQ
  • Entities — tools and concepts, updated as the ecosystem changes
  • Best-of — focused shortlists and decision criteria

How to cite safely

Use the primary source link (official blog/repo/changelog) for any decision or citation. If you cite this site, cite it as a summary layer and follow the source link to verify.

Policy references: Editorial standards · Sources & Coverage · Correction policy

Latest briefs (rolling)

DeepSeek Opens App-Level API, llama.cpp Releases v1.0.0 Stable Version · 1114-825

This week the AI industry reached multiple milestones: DeepSeek fully opened its App-level API capabilities, and llama.cpp released its first stable major version v1.0.0, marking local AI deployment entering maturity [4][16][22]. Meanwhile, Fei-Fei Li's World Labs launched the open version of its world model Atlas, while Cognition and ElevenLabs accelerated their layouts in enterprise agents and voice collaboration respectively [0][10][13].

Review: Editorial review pending

Google Gemini 4 Hydrogen Now Available to Everyone, Tencent Hunyuan Hy4 Tops Cybersecurity Benchmark · Issue 1113-824

Google releases the Gemini 4 Hydrogen model, focusing on speed, cost, and AI Agent capabilities while making it available to everyone [3], and its task automation feature now supports 134 apps [5]; Tencent Hunyuan Hy4 ranks third on the Artificial Analysis Cyber Index and tops CWE-Bench-AA [4]; on the industry side, Jensen Huang resigns as NVIDIA CEO but remains the largest individual shareholder, and Baidu launches the large-model trading platform "Token Exchange" [1...

Review: Editorial review pending

Weekly AI Highlights · 2026-11-13

The model competition has entered a phase of "tiered pricing + specialization": Anthropic splits general-purpose and security/life-science scenarios with the dual Fable 5.2 / Mythos 5.2 line and cuts prices by 20%, while Google pushes Gemini 4 Carbon to 99.1% on ARC-AGI...

Review: Editorial review pending

Google Opens Gemini 4 Hydrogen to All as Open-Source Camp Strikes Back on Two Fronts · 1113-823

Google releases the Gemini 4 Hydrogen model, opening it to all users with faster speed, lower cost, and stronger AI Agent capabilities [0]; meanwhile, the open-source camp continues to push forward, with Perplexity open-sourcing its decision model Decider v1.3 series [1], and Thinking Machines' Inkling 2 Small topping the Code Arena WebDev US open-source model leaderboard with 1647 points [2].

Review: Editorial review pending

Jensen Huang Steps Down as Nvidia CEO, Stock Falls 13%; Google Opens Preorders for $395 Gemini Glasses · 1113-822

Nvidia CEO Jensen Huang has reportedly resigned under pressure from the board, with CFO Colette Kress stepping in as interim head; the news widened the stock's premarket decline from 8% to 13% [9][11][13]. Meanwhile, Google, together with Samsung, Gentle Monster, and Warby Parker, launched Gemini smart glasses starting at $395 and opened preorders [2][6], while ByteDance's Seedream 5.5 went live on the fal platform with text-to-image, editing, and layer separation capabilities, achieving 95% accuracy on small text in complex layouts [3][16][17].

Review: Editorial review pending

Doubao Integrates Kimi and DeepSeek, Baidu Launches Token Exchange · Issue 1112-821

The AI model ecosystem is shifting from single-point competition toward "free combination": Doubao has integrated third-party models such as Kimi and DeepSeek, Baidu has launched the Token Exchange trading platform, and the power to choose model-as-a-service is accelerating toward users [0][2]. Meanwhile, Runway has open-sourced the robotics model Praxis-1 and Volcano Engine has released Seedance 2.5 flash, as the cost-performance race in multimodal and embodied intelligence heats up in parallel [1][6]. On the regulatory side, the U.S. Department of Commerce plans to tighten Nvidia Bl...

Review: Editorial review pending

Google Gemini Decision Pricing: Only $0.1 per Million Tokens · 1112-820

This week the AI industry saw a dense wave of releases: Google launched the Gemini Decision 1 model and API designed specifically for high-speed decision-making, entering routing and scoring scenarios with ultra-low pricing of $0.10 per million input tokens and free output [12][17][21]; meanwhile, Stripe acquired the model aggregation platform OpenRouter for approximately $7.5 billion to $8 billion, marking another blockbuster deal in the AI infrastructure space [1]. On the application side, Runwa...

Review: Editorial review pending

AlphaGenome 2 Doubles Variant Detection Accuracy · 1112-819

Google DeepMind releases the AlphaGenome 2 gene sequence model, nearly doubling variant identification capability and achieving single-cell-level prediction accuracy [1][7][12]; meanwhile, Anthropic is reported to be planning an IPO aiming to raise up to $90.4 billion, potentially becoming the largest listing in history [20], and its model completed an AMD GPU port in just three days, shaking Nvidia CUDA's moat [22]. The enterprise AI battlefield is equally heated, with Databricks Ge...

Review: Editorial review pending

Gemini 4 Carbon Leads, Qwen4-Max · Issue 1111-818

This week the AI industry saw a dual rhythm of dense model releases and accelerated regulatory implementation: Google DeepMind launched Gemini 4 Carbon, its strongest public model to date, followed closely by Alibaba's Qwen4-Max and Baidu's ERNIE 5.5 Flash Preview, with leading vendors continuing to push coding, multi-agent, and long-context capabilities [0][3][10]; meanwhile, New York's RAISE Act regulatory platform opened for registration, bringing OpenAI, Anthropic ...

Review: Editorial review pending

Google Gemini 4 Carbon Achieves 99.1% Reasoning Score · 1111-817

This week, the AI industry saw a dual surge in model capabilities and collaborative ecosystems: Google officially opened its strongest reasoning model, Gemini 4 Carbon, achieving an astonishing 99.1% on the ARC-AGI benchmark [8][22]; meanwhile, Goldman Sachs predicts Anthropic's revenue will exceed $600 billion by 2030 [19]. At the same time, Cursor, Claude, and OpenAI have all ramped up team collaboration and commercialization infrastructure, marking a full shift of AI from a "standalone tool" to an "organization-level..."

Review: Editorial review pending

MiniMax Open-Sources H3-Image: 16-Megapixel Images in 10 Seconds · 1111-816

MiniMax open-sources the H3-Image model, setting a new open-source image generation record with 16-megapixel generation in 10 seconds [2][4][19]; meanwhile, Google DeepMind releases Gemini Robotics 2.5, which leverages Gemini 4 Argon to shorten training cycles by months and achieves 96% sorting accuracy in a Mercedes-Benz factory proof of concept [7][8][9]. On the capital side, Anthropic sets its IPO price range at ...

Review: Editorial review pending

Anthropic's Dual Models Top Intelligence Index as Costs Rise; Apple and Google AI See New Progress on Same Day · 1110-815

Anthropic releases Claude Fable 5.2 and Mythos 5.2, topping the Artificial Analysis Intelligence Index with a score of 65 and planning an IPO, but rising per-task costs draw attention [11][15][13]; meanwhile, Apple expands its China AI filing, Google Gemini launches a memory summary feature, and on the regulatory side, Australia plans to mandate that AI companies report safety violations, with industry and governance advancing on two fronts [1][9][2].

Review: Editorial review pending

Anthropic Releases Claude Fable 5.2, Cracking Black Hole Computation · 1110-814

Anthropic releases Claude Fable 5.2 and Mythos 5.2, achieving breakthroughs on complex tasks such as operating system development and theoretical physics calculations involving black holes, cutting API call fees by 20% and planning an IPO [1][5][7]; meanwhile, GPT-6.2 Sol strikes back on intelligence index, inference speed, and cost, pushing competition among top models into white heat [4]. On the industry side, Jensen Huang's extramarital affair scandal drags in suspicions of an Nvidia espionage investigation, while Meta Muse brings AI agents to wrist-worn devices [6][8].

Review: Editorial review pending

ByteDance Seedream 5.5 Shifts to Pixel-Level Refinement as Chinese Open-Source Models Capture 89% of Enterprise Spending · 1110-813

ByteDance's Seedream 5.5 pushes AI image generation from "gacha-style" luck-based generation toward pixel-level refinement and multi-turn editing, targeting professional design workflows [0][1]; meanwhile, Chinese open-source models hold an astonishing 89% share of global enterprise spending, with Kimi leading at 61% [3]. On the hardware side, Google delivered four eighth-generation TPU variants within a year and introduced AMD EPYC as a host processor for the first time, accelerating competition with Nvidia [11][24].

Review: Editorial review pending

NVIDIA AI Day Seoul: Samsung and Hyundai Partnership Speeds Up Semiconductor Simulation 1,600x · 1109-812

NVIDIA held AI Day in Seoul, rolling out physical AI and sovereign AI collaborations with the South Korean government and giants including Samsung, LG, and Hyundai, with a semiconductor simulation case achieving roughly 1,600x acceleration [8][17][18][21]. Meanwhile, Microsoft CEO Nadella proposed an organizational vision of "20 million agents collaborating with 200,000 employees" [5], while a leaked Google internal memo admitted there is no moat against OpenAI [6]. On the application side, WeChat Input Method launched a free AI voice recorder [0], and Cl...

Review: Editorial review pending

AlphaEvolve Makes Benchmarks Harder Through Evolution · 1109-811

Stanford and Google DeepMind have teamed up to use AlphaEvolve to "evolve" static tool-calling benchmarks to be harder, directly targeting the industry pain point that evaluations are too easy [1]; meanwhile, an Amazon audit found that the reference answers and scoring code of 9 policy decision benchmarks contain errors themselves [17], and a mysterious new Anthropic model doubled its RSI speed in a METR investigation [21]—evaluation credibility and model capabilities are being interrogated in tandem.

Review: Editorial review pending

MCP Interactive Tutorial and Scene Theme Tool Debut, Developer Efficiency Chain Evolves Again · Issue 1109-810

This issue of the tech weekly focuses on the deep integration of AI toolchains and development efficiency, from the MCP interactive tutorial to the Scene theme unification tool, and the lightweight macOS virtual machine Vimy, showcasing the developer ecosystem's continuous exploration of lightweight solutions and cross-platform collaboration [1]. At the same time, sharing of AI product design methods and personal life practices reflects the tech community's deeper thinking on human-machine collaboration and life balance [1].

Review: Editorial review pending

Fujifilm Expands Tape Production Fivefold, Behind OpenAI's $614 Billion Compute Spending · Issue 1108-809

The explosion of generative AI is reshaping the entire industry chain from underlying storage to upper-layer applications: Fujifilm is expanding tape production fivefold due to AI data archiving demand [0], while OpenAI's compute infrastructure spending through 2030 is estimated to reach as high as $614 billion [3]. Meanwhile, Chinese open-source models have reached 89% of global enterprise spending, with Kimi, DeepSeek, and GLM ranking in the top three [8], but capital is becoming more cautious about the world model track [17].

Review: Editorial review pending

Anthropic Races Toward IPO Before Thanksgiving, Valued at Up to $1.75 Trillion · Issue 1108-808

Anthropic plans to launch its IPO before Thanksgiving, aiming to raise $80–100 billion at a valuation of $1.25–1.75 trillion, marking a landmark event in AI industry capitalization [18]. Meanwhile, New York State Assemblymember Alex Bores is calling on the U.S. Congress to legislate a ban on recursive self-improvement (RSI), further highlighting the tension between AI safety governance and commercial expansion [1]. On the technical side, Google has donated agent runtime infrastructure to the CNCF, and Cerebras demonstrated...

Review: Editorial review pending

Anthropic Model Improvement Speed Doubles, Triggering Safety Warnings · Issue 1108-807

AI safety is facing a dual challenge: on one hand, the doubling of Anthropic's model improvement speed has raised concerns among safety researchers, with Alex Stamos warning that the industry will enter a "dark age" of cyberattacks [17][0]; on the other hand, a new benchmark from a CMU team shows that none of 9 AI agents passed expert review on open-ended research tasks, exposing systemic shortcomings in honesty and completeness among current agents [24].

Review: Editorial review pending

Weekly synthesis

If you want a higher-level view (patterns and decisions), use the weekly report.