Models

Fireworks 实测:MiniMax M3.1 Flash 编程表现突出,成本仅 Gemini 3.8 Flash 的零头

Why we picked it

做编程类 Agent 的可以看看这份实测:MiniMax M3.1 Flash 跑 DeepSWE 任务比 Gemini 3.8 Flash 还多,成本低 92%。更有用的是 Fireworks 给了切换、重试、多模型组合的具体策略,选型和省钱都用得上。

Fireworks AI 发布基于 SII 模型的新研究,用从业者自建基准评测 MiniMax M3.1 Flash 在编程任务上的表现。结果显示 M3.1 Flash 完成的 DeepSWE 任务数量超过 Gemini 3.8 Flash,同时模型成本低 92%。

Image · Fireworks AI
Body

这项研究的核心是用真实开发者构建的基准而非合成测试,来衡量不同模型在质量与成本之间的取舍,并给出具体策略:什么时候该切换模型、什么时候该重试、什么时候该组合使用多个模型。

对开发者而言,M3.1 Flash 以极低成本拿到超过 Gemini 3.8 Flash 的编程完成率,意味着大规模跑代码类 Agent 任务时的单位成本可能明显下降;企业选型时也可以把"切换/重试/组合"策略纳入路由设计,而不是绑定单一模型。

Original · Fireworks AI

M3.1 Flash from @MiniMax_AI stands out in coding. It completes more DeepSWE tasks than Gemini 3.8 Flash at 92% lower model cost. Our new SII model study uses practitioner-built benchmarks to examine where it improves quality and cost, and when to switch, retry, or combine models. Read our new blog and dig into the benchmarks: bit.ly/minimaxsiix

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Models" desk, focusing on China's AI industry and domestic foundation model ecosystem. Within our cross-regional analytical framework, it directly maps to US-China foundation model benchmarks and commercialization comparison and AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include MiniMax, M3.1 Flash, Gemini 3.8 Flash, coding assistants.

Source attribution: curated from Fireworks AI (@FireworksAI_HQ).