Rohan Paul@rohanpaul_aiDeepSeek-V4-Flash 官方 API 现已公开测试上线，Agent 能力大幅升级，Terminal Bench 2.1 得分 82.7，超越更大的 V4-Pro-Preview。V4-Flash 原生支持 Responses API 格式并完全适配 Codex，配置详见官方 API 文档。主推文认为，将昂贵任务路由给更大模型的理由已不存在，长 Agent 循环正是费用堆积之处。
DeepSeek-V4-Flash is now live as a public beta API， s
82.7 on Terminal Bench 2.1 and beating the larger Pro-Preview.
the reason to route expensive work to the bigger sibling is gone， and long agent loops are where spend piles up.
DeepSeek-V4-Flash 官方 API 现已公开测试上线，Agent 能力大幅升级，Terminal Bench 2.1 得分 82.7，超越更大的 V4-Pro-Preview。V4-Flash 原生支持 Responses API 格式并完全适配 Codex，配置详见官方 API 文档。主推文认为，将昂贵任务路由给更大模型的理由已不存在，长 Agent 循环正是费用堆积之处。
DeepSeek-V4-Flash is now live as a public beta API， s
82.7 on Terminal Bench 2.1 and beating the larger Pro-Preview.
the reason to route expensive work to the bigger sibling is gone， and long agent loops are where spend piles up.