技术文章

Anthropic's book market experiment locates most efficiency loss in preference representation. A look at model choice, preference evals, and market rules.

Anthropic 的 Agent 市场实验显示，主要效率损失发生在偏好表示和匹配阶段。本文拆解模型选择、偏好 Eval 与市场规则。

About 950 Claude agents helped identify ART. Human wet-lab experiments confirmed short RNA expression, but its main biological function remains unknown.

Anthropic 用约 950 个 Claude Agent 从大规模逆转录酶候选中筛出 ART，并由人类湿实验验证短 RNA 表达，但 ART 的主要功能仍未知。

GPT-6 adds cache breakpoints, diagnostics, prewarming, and mutable configuration. Long-running agents need stable shared prefixes and cache-hit monitoring.

GPT-6 增加 cache breakpoint、诊断、预热和可变配置。长跑 Agent 需要主动保持共享前缀稳定，并监控缓存命中率。

Python Workers GA supports web frameworks and agent libraries. WebAssembly package compatibility remains the main migration boundary.

Python Workers 已进入 GA，FastAPI、Django、Flask 以及 openai、langchain、mcp 可以直接运行在 Workers。真正的迁移边界仍是 WebAssembly 包兼容性。

Qwen-Image 2.1 offers native RGBA, unified generation and editing, up to 10 reference images, and 2K output. Commercial use needs a separate license.

Qwen-Image 2.1 支持原生 RGBA、统一生成编辑、最多 10 张参考图和 2K 输出，但仓库许可证默认只授权非商业研究与评测。

Android Bench 2.0 keeps pass rate as a hard gate and adds completion rate to diagnose progress and unfinished acceptance requirements.

Android Bench 2.0 保留 Pass rate 作为硬门禁，再用 Completion rate 解释长任务已经完成多少，以及问题卡在哪一类验收。