#OSWorld benchmark
13 related articles
·5 min
OSWorld 2.0 Benchmark: Top AI Agent Achieves Only 20% Completion Rate — Long-Horizon Tasks Are the Achilles' Heel
OSWorld 2.0 benchmark tests 108 long-horizon computer tasks. Claude Opus tops at only 20.6% completion, exposing critical AI weaknesses in state tracking and error self-correction.
Read more →
·5 min
OSWorld 2.0 Benchmark: Top AI Agents Achieve Only 20% Completion Rate — Long-Horizon Tasks Are the Fatal Weakness
OSWorld 2.0 benchmark tests 108 long-horizon computer tasks (median 1.6 hrs for humans). Claude Opus tops out at 20.6% completion, exposing critical AI Agent weaknesses in state maintenance and self-correction.
Read more →
·3 min
Zhipu's Mobile AI Agent Tested: A Full Breakdown of the Cloud Phone Approach — Capabilities and Limitations
Hands-on test of Zhipu's mobile AI Agent: using a cloud phone to bypass permission limits, it supports natural language-driven automation. We cover its core mechanics, real performance, app restrictions, and future potential.
Read more →
·6 min
GPT-5.6 Hands-On Deep Dive: ChatGPT and Codex Merge, Full Analysis of the Sol Model
GPT-5.6 is officially released, merging ChatGPT and Codex into one app and launching the three-tier Sol, Terra, and Luna models. A detailed breakdown of 16 hands-on tests plus Worker mode and Codex dev upgrades.
Read more →
·6 min
GPT-5.6 Fully Available: A Deep Dive into the Sol/Terra/Luna Three-Tier Models and Four-Agent Parallelism
GPT-5.6 is now officially available to all users, launching the three-tier Sol, Terra, and Luna models with four-agent parallelism. An in-depth look at the official benchmarks, API pricing, safety, and Ultra mode.
Read more →
·4 min
Reinforcement Learning for AI Agents: A Deep Dive into RLHF and Agentic RL
A deep dive into RL for AI agents: from RLHF to Agentic RL, covering PPO vs. GRPO, sparse rewards, tool-calling optimization, and verifiable rewards.
Read more →
·4 min
Agent Draw: An AI Whiteboard That Draws While You Talk
Agent Draw is an AI whiteboard built on TLDraw that lets you speak or type to have an AI agent draw flowcharts and diagrams in real time. A deep dive into its tech, design, and use cases.
Read more →
·4 min
Agent Tool Selection Guide: A Comprehensive Comparison of API, CLI, MCP, Browser Use, and Computer Use
Compare 5 Agent tool types: CLI, API, MCP, Browser Use & Computer Use on speed, accuracy, and token cost. Includes a selection priority table to cut costs and boost Agent efficiency.
Read more →
·2 min
Google Gemini 3.5 Flash Released: A Generational Leap Focused on Agentic and Coding Capabilities
Google releases Gemini 3.5 Flash, skipping version 3.0 in a generational leap focused on agentic capabilities and coding performance, positioning it as a new AI model family bridging frontier intelligence with real-world action.
Read more →
Tech Frontiers·1 min
GPT-5.4 In-Depth Review: Native Computer Use, Unified Reasoning & Coding — OpenAI Reclaims the Throne
GPT-5.4 full review: Surpasses Claude Opus 4.6 on OSWorld, native computer use, 50% better token efficiency in reasoning+coding, 33% fewer hallucinations, and record-breaking search. OpenAI's first all-in-one model.
Read more →
Deep Dives·3 min
Auto-Use Open Source Project Explained: One AI Agent to Handle Your OS, Browser, and Code
Deep dive into the Auto-Use open source project—an end-to-end AI Computer Use Agent that unifies OS control, browser automation, and coding. Analysis of architecture, competitors, and industry trends.
Read more →
Deep Dives·2 min
computer-use-windows: An Open-Source Framework for Building AI Desktop Agents on Windows
An open-source project based on Anthropic's Computer Use API, computer-use-windows enables AI agents to directly control the Windows desktop for automation tasks including office workflows and software testing.
Read more →
