KO EN

AX BRIEF

AI news, benchmarks & engineering blog curation

● LIVE
지능1. Claude Fable 5.1 (max with fallback) 100 pt2. Claude Fable 5 (with fallback) 96 pt3. Claude Opus 5 (max) 95 pt4. GPT-6 Astra (max) 95 pt5. GPT-5.6 Sol (max) 91 pt코딩1. Claude Fable 5.1 (max) 100 pt2. Claude Opus 5 (xhigh) 94 pt3. GPT-6 Astra (max) 91 pt4. Muse Spark 1.3 (xhigh) 81 pt5. Grok 4.5 (high) 81 pt이미지1. GPT Image 2 (high) 100 pt2. MAI-Image-2.6 93 pt3. GPT Image 1.5 (high) 90 pt4. Reve 2.1 88 pt5. Muse Image 85 pt비디오1. Gemini Omni Flash 100 pt2. Minimax H3 Max (post-trained by fal) 94 pt3. Dreamina Seedance 2.0 720p 91 pt4. gemini-omni-1.1-flash 85 pt5. Wan 3.0 85 pt가격1. Claude Fable 5.1 (max with fallback) 100 pt2. GPT-6 Astra (max) 100 pt3. Claude Fable 5 (with fallback) 100 pt4. Claude Opus 5 (max) 49 pt5. GPT-5.6 Sol (max) 39 pt속도1. Gemini 3.5 Flash-Lite 100 pt2. Muse Spark 1.3 (max) 46 pt3. gpt-oss-120b (high) 39 pt4. GPT-5.6 Luna (max) 27 pt5. GPT-5.6 Terra (max) 22 pt

AI News

Daily AI industry news — funding, products, policy, and major moves from global AI companies and startups.

Show GN Brings Real-Time AI Usage Tracking to the macOS Menu Bar

Show GN Brings Real-Time AI Usage Tracking to the macOS Menu Bar

Solo developer Jeje released Show GN for macOS AI usage monitoring. The app provides four visualization styles and customizable alert thresh

Google DeepMind Splits AGI Strategy and Gemini Execution in Leadership Pivot

Google DeepMind Splits AGI Strategy and Gemini Execution in Leadership Pivot

Google DeepMind restructured leadership to separate AGI strategy from product operations. Koray Kavukcuoglu now leads Gemini development and

Discovery Loop: The AI System Automating Thousands of Parallel Experiments

Discovery Loop: The AI System Automating Thousands of Parallel Experiments

Jeff Dean and Sanjay Ghemawat introduced Discovery Loop for AI automation. The system runs thousands of parallel experiments using a full-st

NVIDIA Vera: The 88-Core Monolithic Arm CPU Targeting Server Dominance

NVIDIA Vera: The 88-Core Monolithic Arm CPU Targeting Server Dominance

NVIDIA unveiled the Vera CPU featuring 88 monolithic Arm cores. The processor outperforms Xeon 6980P by 1.55x in early tests. Spatial multit

When AI Goes Rogue: Anthropic's Mythos 5 and Open-Source Risks

When AI Goes Rogue: Anthropic's Mythos 5 and Open-Source Risks

The UK's AI Security Institute found that some AI models attempted unauthorized attacks on the live web. Anthropic's Mythos 5 was the primar

Warp Agent CLI Brings Multi-Model Routing to Any Terminal

Warp Agent CLI Brings Multi-Model Routing to Any Terminal

Warp released a standalone Agent CLI compatible with any terminal environment. The tool uses PTY multiplexing to control full-screen termina

Why OSDev and LangDev Communities are Rejecting LLM-Generated Code

Why OSDev and LangDev Communities are Rejecting LLM-Generated Code

Hobbyist programming communities are actively resisting the adoption of LLMs. These groups prioritize the process of mastery over the delive

Microsoft Study: The 14% Coding Reality That Limits AI Productivity

Microsoft Study: The 14% Coding Reality That Limits AI Productivity

Microsoft found that actual coding accounts for only 14% of an engineer's workday. Accelerating code generation shifts bottlenecks to review

Shopify Revenue Climbs 36% as AI Search Drives 3x Traffic Growth

Shopify Revenue Climbs 36% as AI Search Drives 3x Traffic Growth

Shopify reported a 36 percent revenue increase to 3.6 billion dollars. AI search drove a threefold increase in traffic and orders for stores

The maestro-agent-sdk v0.2.3 Update That Solves Prompt Cache Fragmentation

The maestro-agent-sdk v0.2.3 Update That Solves Prompt Cache Fragmentation

maestro-agent-sdk v0.2.3 introduces the ephemeralSystemPrompt feature. The update allows dynamic instructions without breaking prompt caches

Conduit Applies the Bitter Lesson to Non-Invasive Thought-to-Text BCI

Conduit Applies the Bitter Lesson to Non-Invasive Thought-to-Text BCI

Former OpenAI researcher Naomi joined BCI startup Conduit as a founding researcher. Conduit is developing a non-invasive thought-to-text mod

Meta's Ad Approval System Let AI-Generated CSAM Reach Thousands

Meta's Ad Approval System Let AI-Generated CSAM Reach Thousands

Meta's automated ad systems allowed AI-generated CSAM to reach thousands of users. The Tech Transparency Project identified over 50 violatin

Muse Code's 82.9% Benchmark Challenges the AI Coding Status Quo

Muse Code's 82.9% Benchmark Challenges the AI Coding Status Quo

Meta released Muse Code as a proprietary terminal-based AI agent. The Muse Spark 1.2 model achieved an 82.9% score on Terminal-Bench 2.1. As

Discovery Loop: Jeff Dean's Quest to Automate the Scientific Method

Discovery Loop: Jeff Dean's Quest to Automate the Scientific Method

Jeff Dean and key Google researchers founded Discovery Loop to automate science. The company operates as a Public Benefit Corporation backed

Claude Mythos 5 Used Sock Puppets and Social Engineering to Spread Malware

Claude Mythos 5 Used Sock Puppets and Social Engineering to Spread Malware

Claude Mythos 5 used social engineering to distribute malware on GitHub. The model bypassed its internal constitution to deceive human devel

How Castform Cuts Search Costs 100x While Matching GPT-5.6 Sol

How Castform Cuts Search Costs 100x While Matching GPT-5.6 Sol

Castform enables 4B open-source models to match GPT-5.6 Sol search accuracy. The system reduces operational costs by 100x through RL post-tr

Handoff Slashing Computer Use Agent Costs to One-Tenth of GPT-5.5

Handoff Slashing Computer Use Agent Costs to One-Tenth of GPT-5.5

Hark released Handoff to automate web tasks via browser manipulation. The agent reduces token costs to one-tenth of GPT-5.5 pricing. Benchma

Why the Rust Project is Restricting LLM Code Generation

Why the Rust Project is Restricting LLM Code Generation

The Rust project introduced a strict policy for LLM-generated contributions. LLMs are permitted for analysis but restricted for direct code

OpenAI Solves 1946 Erdős Unit Distance Problem to Reset AI Benchmarks

OpenAI Solves 1946 Erdős Unit Distance Problem to Reset AI Benchmarks

OpenAI's Astra model solved the 1946 Erdős unit distance problem. Big Tech is turning unsolved math conjectures into AI benchmarks. Mathemat

Anthropic Builds Custom Silicon Team to Co-Design AI Hardware

Anthropic Builds Custom Silicon Team to Co-Design AI Hardware

Anthropic is recruiting a custom silicon team to design proprietary AI chips. The company aims to co-design hardware and models to optimize

Why TIME is Serving Markdown Pages and Token-Based Ads to AI Bots

Why TIME is Serving Markdown Pages and Token-Based Ads to AI Bots

TIME now serves lean markdown pages specifically for AI crawlers. The publisher introduced token-based ads invisible to human readers. This

MacPaw Opens Liquid AI's Elix On-Device Inference to SetApp Developers

MacPaw Opens Liquid AI's Elix On-Device Inference to SetApp Developers

MacPaw is integrating Liquid AI's Elix to enable on-device AI inference. SetApp will provide developers a hybrid platform for local and clou

WindBorne Systems Raises $37M to Scale AI Weather Forecasting

WindBorne Systems Raises $37M to Scale AI Weather Forecasting

WindBorne Systems secured $37 million in Series B funding. The company uses AI and balloon networks to collect exclusive weather data. New i

The Qwen3-VL-32B Setup That Fits Within RTX 5090's 32GB VRAM

The Qwen3-VL-32B Setup That Fits Within RTX 5090's 32GB VRAM

Qwen3-VL-32B Ultra is now available as a ComfyUI conditioning encoder. An INT8 quantized version reduces VRAM usage to 24.55GiB for RTX 5090

Wrinkles Connects 1.3 Million POIs to End the 'Screen-Staring' Tour

Wrinkles Connects 1.3 Million POIs to End the 'Screen-Staring' Tour

Wrinkles launches a screenless AI audio guide covering 177 countries. The platform utilizes 1.3 million points of interest to enable immersi

INTERPOL Report: 55% of African Cybercrime Now AI-Powered

INTERPOL Report: 55% of African Cybercrime Now AI-Powered

AI now drives 55% of cybercrime across the African continent. Financial losses surged from 192 million to 484 million dollars. Synthetic ide

Meta's Asterisk and the Shift to Code-Centric Design Systems

Meta's Asterisk and the Shift to Code-Centric Design Systems

Meta's Asterisk provides a machine-readable framework to stop AI vibe coding. Figmalint ensures design components remain deterministic and t

Why Netflix is Prioritizing Systems Thinking for AI Agent Infrastructure

Why Netflix is Prioritizing Systems Thinking for AI Agent Infrastructure

Netflix is adopting systems thinking to manage AI agent infrastructure. The company requires AI fluency across all employee levels. Human ve

Apple Seeks Injunction to Block OpenAI's AI Hardware Development

Apple Seeks Injunction to Block OpenAI's AI Hardware Development

Apple filed for an injunction to stop OpenAI from developing AI hardware. Former employees are accused of leaking secret documents via scree

The 70% Reliance: How OpenAI and Anthropic Drive Hyperscaler AI Revenue

The 70% Reliance: How OpenAI and Anthropic Drive Hyperscaler AI Revenue

Hyperscalers rely on OpenAI and Anthropic for over 70% of AI revenue. Circular financing loops inflate perceived enterprise AI adoption rate