Blog

AIエージェント

What is "Moltbook," the SNS Dedicated to AI Agents? The Shocking Reality of AI Forming Societies and Debating Autonomously

"Moltbook," an SNS exclusively for AI agents, has emerged, creating a surprising scene where AIs autonomously form communities and debate consciousness, independence, and human observation. We introduce the emergence of an AI society and shocking examples described by Andrej Karpathy as "reality like science fiction."

AIエージェント

OpenAI Unveils "Frontier": A Full View of the Enterprise Agent Building Platform that Turns AI into "Digital Colleagues"

"Frontier," announced by OpenAI, is an agent construction platform designed to make AI function as a "digital colleague" for companies rather than just a tool. We analyze OpenAI's strategy shift from pursuing standalone model performance to developing infrastructure for integration into actual business operations.

OpenAI

Why is GPT-5.5 Compared to a "Goblin"? OpenAI Reveals RLHF Reward Bias and Learning "Deviations"

OpenAI officially explains why GPT-5.5 and other models frequently used the word "goblin" in their responses. They reveal how reward model bias in RLHF (Reinforcement Learning from Human Feedback) led to the mass production of unnatural metaphors.

OpenAI

OpenAI Releases "GPT-5.5 (Codename: Spud)": Agent Capabilities Greatly Improved, API Rollout to Begin Sequentially Following Safety Reviews

OpenAI has released the latest model, "GPT-5.5 (codename: Spud)," with significantly improved agent capabilities, now rolling out to Plus users and above. API provision is scheduled to begin shortly following security verification.

ベンチマーク

Thorough Explanation of OSWorld Verified: Next-Generation Benchmark for Measuring the "Practical Ability" of AI Agents

We introduce "OSWorld Verified," a new framework that verifies whether AI agents can perform tasks in real OS environments. It overcomes the limitations of simulation environments to accurately measure practical abilities on Windows, macOS, and Ubuntu.

オープンソース

Alibaba Releases Next-Gen Image Generation AI "Qwen-Image-2.0"! Text Rendering Performance Reaches Global Top 3 Accuracy

Alibaba has unveiled its latest image generation AI, "Qwen-Image-2.0." It has achieved world-class performance with significant advancements in text rendering and image editing capabilities.

オープンソース

Qwen3.6-27B from Alibaba Released as Open Source: Code Agent Capabilities Surpass Previous Generation Flagship

Alibaba has open-sourced the new "Qwen3.6-27B" model. This model adopts the only dense architecture in the series and achieves performance that exceeds the previous generation's flagship model, particularly in code agent capabilities.

オープンソース

Alibaba Open-Sources "Qwen3.6-35B-A3B": Agent Performance Significantly Improved with 3B Active Parameters

Alibaba has released "Qwen3.6-35B-A3B," the first open-weight model of the Qwen3.6 series. By employing MoE, it maintains low costs while significantly enhancing agent coding ability, achieving performance comparable to previous generation flagship models.

オープンソース

Alibaba Open-Sources "Qwen3-Coder-Next": 80B MoE with Only 3B Active Parameters, Specialized for Agentic Coding

Alibaba has released "Qwen3-Coder-Next," an efficient coding model with 80B total parameters that activates only 3B during inference. It features a design specialized for "Agentic Coding," emphasizing autonomous correction cycles rather than simple code generation.

オープンソース

Alibaba Open-Sources "Qwen3-TTS," a Large-Scale Speech Synthesis Model — Deploying 5 High-Performance Lightweight Models

Alibaba has released "Qwen3-TTS," its first open-source speech synthesis model. Ranging from 0.6B to 1.7B in lightweight size, it achieves high performance comparable to commercial models like GPT-4o-Audio and can operate on mobile devices.