Four hours, 672 tool calls: MiMo-V2.5-Pro's long-horizon run logs

MiMo-V2.5-Pro is built for long-horizon, high-difficulty tasks. On a real compiler project from Peking University, one of China’s top universities, MiMo-V2.5-Pro autonomously built a complete SysY compiler in Rust from scratch — including lexer, parser, AST, Koopa IR codegen, RISC-V backend, and performance optimization. A project that typically takes PKU computer science students several weeks was completed in just 4.3 hours. Across 672 tool calls, it achieved a perfect 233/233 on the hidden test suite. Even its first compile passed 137/233 tests — a 59% cold-start pass rate, pointing to a strong overall design before iterative debugging began.

With just a few prompts, it autonomously delivered a working desktop video editor — complete with a multi-track timeline, clip trimming, cross-fades, audio mixing, and an export pipeline. The final build spans 8,192 lines of code, produced over 1,868 tool calls across 11.5 hours of autonomous work. https://t.co/LDVxxi1jgA

On a graduate-level analog EDA task, it designed and optimized a complete FVF-LDO from scratch in the TSMC 180nm CMOS process — sizing the power transistor, tuning the compensation network, and choosing bias voltages so that six key metrics all landed within spec: phase margin, line regulation, load regulation, quiescent current, PSRR, and transient response. A trained analog designer would typically spend several days on a project like this. Wired into an ngspice simulation loop with Claude Code as the harness, MiMo-V2.5-Pro reached a fully spec-compliant design in about 1 hour. The four metrics shown below were improved by an order of magnitude over its own initial attempt.

Let's build with MiMo V2.5! https://t.co/4oM2kMGAZS