The author explains how OpenAI's GPT-6 model family with computer-use capabilities and halved pricing eliminates the need for tools like Navicat and reduces Spring to a template, shifting backend developers from code writing to problem definition and model output auditing.
The article analyzes B站's AI Arena evaluation where GPT-6 Astra topped the leaderboard, but reveals its victory is limited to agent execution and code repair tasks, while other models excel in different real-world scenarios, exposing the gap between standardized benchmarks and practical performance.
OpenAI's GPT-6 Astra integrates reasoning, computer operation, and continuous execution into a single model, scoring 72.6% on OSWorld 2.0, 57.9% on Terminal-Bench 4.0, and 99.9% on ARC-AGI-3 with a Provider Adapter, while demonstrating autonomous tax filing, CRM updates, code migration, and zero-day vulnerability discovery — all with new cross-context memory and safety boundaries.
The article explains how Agentic Reinforcement Learning workloads like OSWorld need an execution environment that preserves state across hundreds of actions, and shows that integrating Harbor with Alibaba Cloud Function Compute sandbox meets four strict requirements, enables full‑scale evaluation, and reduces compute‑only costs by about 60%.
Amap's open-source LongHorizon-Harness wraps existing agents (Claude Code, Codex, OpenCode, DeepSeek Harness) in a Manager-Executor-Auditor loop, raising WeaveBench pass rate from 51.8% to 80.7%, tripling OSWorld 2.0 desktop task completion, and cutting Terminal-Bench 2.1 token use by 24% — all without changing the underlying models.
EvoCUA, a native computer‑use agent from Meituan, combines a verifiable data‑synthesis engine, a ten‑thousand‑level sandbox infrastructure, and an experience‑driven learning paradigm to overcome data‑scaling and feedback challenges, achieving a 56.7% success rate on the OSWorld benchmark and surpassing previous open‑source models.
