做浏览器自动化或代理系统的团队，这个对比直接告诉你模型选择如何影响生产环境的成本和稳定性——Kimi/GLM/MiniMax 的低重试率值得关注。
Fireworks AI 与 NotteCore 合作，在多个前沿模型上运行了 720 个浏览器代理任务。结果显示，某个基线模型在约 1/5 的调用中产生格式错误输出，导致多步工作流中频繁重试。而 Kimi K2.5、GLM-5 和 MiniMax M2.5 在 Fireworks 上运行时，重试率近乎为零，且随着任务步骤增加，延迟保持稳定。这一差异在生产级代理系统中直接体现为成本、延迟和可靠性的分化。完整报告已发布。
We ran 720 browser agent tasks with @nottecore across frontier models. One baseline model produced...
We ran 720 browser agent tasks with @nottecore across frontier models. One baseline model produced malformed outputs in ~1 out of every 5 calls, leading to retries inside multi-step workflows. Across Kimi K2.5, GLM-5, and MiniMax M2.5 served on Fireworks, retry rates were near zero and latency stayed stable even as tasks extended across multiple steps. Same workload. Same agent loop. Different execution behavior. That gap is what shows up as cost, latency, and reliability divergence in production agent systems. Read the report: fireworks.ai/blog/agent-exe… 💬 1 🔄 1 ❤️ 11 👀 437 📊 3 ⚡