Xiaomi MiMo Code outperforms Claude Code in complex 200+ step tasks, showcasing advanced AI capabilities with free MiMo Auto

Editorial illustration for Xiaomi's MiMo Code beats Claude Code on 200+ step tasks, free MiMo Auto to V2.5

Xiaomi's MiMo Code beats Claude Code on 200+ step tasks,...

Xiaomi's MiMo Code beats Claude Code on 200+ step tasks, free MiMo Auto to V2.5

By AI Daily Post Edited by Brian Petersen, Editor-in-Chief

June 12, 2026 • 3 min read

Here's the thing: Xiaomi just dropped MiMi Code, an open‑source coding assistant that claims to outpace Anthropic’s Claude Code on tasks that stretch beyond 200 steps. The bundle includes MiMo Auto, a zero‑configuration channel that gives developers free, limited‑time access to MiMo‑V2.5—Xiaomi’s multimodal model released in late April 2026. That model packs 310 billion parameters in total, but only 15 billion fire up per inference, and it can handle a 1 million‑token context window, a spec the company says matches Claude Sonnet 4.6 for multimodal agentic work.

While the tech is impressive, the real lure may be the price. MiMo‑V2.5 starts at $0.40 per million input tokens and $2.00 per million output tokens; the larger V2.5‑Pro, a 1.02‑trillion‑parameter mixture‑of‑experts model with 42 billion active parameters, runs $1.00/$3.00 per million up to a 256K context, dropping to as low as $0.20–$0.40 with cache hits. VentureBeat noted the MIT‑licensed models are among the most efficient and affordable for agentic tasks, and the Pro version’s “harness awareness” training aims to manage its own memory within frameworks like Claude Code or OpenCode.

MiMo Code ships with "MiMo Auto," a zero-configuration channel offering free, limited-time access to MiMo-V2.5 -- the natively multimodal model Xiaomi released in late April 2026, a sparse mixture-of-experts design with 310 billion total parameters (just 15 billion active per inference) and a 1 million token context window, which the company positions as matching Anthropic's Claude Sonnet 4.6 in multimodal agentic work.
As VentureBeat reported when the MiMo-V2.5 family launched in April, the models are MIT-licensed and among the most efficient and affordable available for agentic tasks.
The larger MiMo-V2.5-Pro -- a 1.02-trillion-parameter mixture-of-experts model with 42 billion active parameters and a hybrid-attention architecture -- led the open-source field on Xiaomi's ClawEval agentic benchmark with a 63.8% success rate while consuming only about 70,000 tokens per trajectory, roughly 40-60% fewer than Anthropic's Claude Opus 4.6, Google's Gemini 3.1 Pro, or OpenAI's GPT-5.4 needed for comparable results.

Xiaomi's new open source, agentic AI coding harness MiMo Code beats Claude Code at ultra-long, 200+ step tasks - VentureBeat AI

Why this matters We see Xiaomi pushing an open‑source, agentic AI coding tool that outperformed Claude Code on tasks exceeding two hundred steps, a benchmark that suggests the system can handle unusually long reasoning chains. The bundle includes MiMo Auto, a zero‑configuration channel that grants free, limited‑time access to MiMo‑V2.5, Xiaomi’s multimodal model introduced in late April 2026. That model uses a sparse mixture‑of‑experts architecture with 310 billion total parameters but activates only 15 billion per inference, and it supports a one‑million‑token context window.

For developers, the promise of such scale without full‑cost compute could lower entry barriers, especially when the code assistant is freely available. Yet the performance gap is demonstrated only on ultra‑long, 200‑plus step tasks; it is unclear whether similar gains appear on more typical coding workloads or across diverse programming languages. Founders may appreciate the aggressive pricing and bundled access, but the sustainability of free limited‑time usage and the real‑world robustness of the model remain open questions.

Researchers will likely watch how the sparse expert activation balances cost and capability as the ecosystem evolves.