跳到主内容
@wquguru
精选78Rohan Paul模型发布/更新

Claude Opus 5在SWE-bench得分上近乎三倍于Qwen模型

Claude Opus 5 nearly tripled a Qwen model's SWE-bench Verified score while worki…

原文
发到 X

Claude Opus 5 nearly tripled a Qwen model's SWE-bench Verified score while working as an automated AI researcher.

Claude Opus 5 在作为自动化 AI 研究员工作时,将 Qwen 模型的 SWE-bench Verified 分数几乎提升了三倍。

@Evolvent_AI 's newly released open-source RSIGym made that measurement possible by letting an AI agent retrain a model and rewrite its harness inside one budgeted environment.

@Evolvent_AI 新发布的开源工具 RSIGym 使得该测量成为可能,它允许 AI 代理在一个预算受限的环境中重新训练模型并重写其测试框架(harness)。

shows that frontier AI agents can substantially improve another AI model when training, serving, and testing come as ready-made services.

研究表明,当训练、部署和测试都以现成服务的形式提供时,前沿 AI 代理能够显著提升另一个 AI 模型的性能。

Agents work from CPU-only containers and call remote services for LoRA fine-tuning, model serving, benchmarking and sandboxes, all charged against a per-run budget.

这些代理运行在仅含 CPU 的容器中,并调用远程服务进行 LoRA 微调、模型部署、基准测试和沙箱操作,所有费用均从每次运行的预算中扣除。

In the main test, 6 frontier agents started from Qwen3.5-35B-A3B-Base and a minimal harness, with $500 of services per benchmark.

在主测试中,6 个前沿代理从 Qwen3.5-35B-A3B-Base 模型和一个最小化的测试框架开始工作,每个基准测试配备 500 美元的服务预算。

If you're improving an agent, read its failure logs and fix the harness first: heavier training scored lower in 8 of 10 small tests.

如果你正在优化一个代理,请先阅读其失败日志并修复测试框架:在 10 项小型测试中的 8 项里,更重的训练策略得分反而更低。

更进一步:量化金融体系

看懂新闻只是起点——沿量化金融路径,把它变成能交付的工程能力

进入量化体系 →

关联讨论

同一事件的更多信源

相似阅读

关联信息,但可能不是同一事件