If you can't get access to Claude Opus or GPT-5 but still want to set up a usable AI coding environment locally, Xiaomi's newly released MiMO-V2.5-Pro is one of the most cost-effective options available right now. It ties with Kimi K2.6 for first place among open-source models on the AA leaderboard, supports 1M-token context, has extremely low API latency, and costs only 40% of Opus's price. Even better -- it works with Claude Code beautifully smoothly.

Why This Combination

• Cheap: MiMo-V2.5-Pro API pricing is 7/21 RMB per million tokens (input/output, 0-256K), roughly 60% cheaper than Opus 4.6 ($5/$25 per million tokens)
• Fast: it just launched and has few users, so API latency is extremely low -- in real-world tests it beats many Chinese models
• Capable: tool calling and coding ability feel close to Opus 4.6 -- it "speaks plainly" and its output is well organized
• 1M context: a context window in the same class as GPT-5.5

Preparation Before You Start

You need to prepare the following:

• Claude Code: installed and configured (if you haven't installed it yet, see this step-by-step tutorial)
• cc-switch: a model-switching tool for Claude Code, used to hook up third-party models
• Xiaomi API key: apply for one on the Xiaomi Open Platform
• Estimated time: 10 minutes

Step-by-Step Instructions

Step 1: Get a Xiaomi API Key

Head to the Xiaomi Open Platform, register, and apply for an API key for the MiMo models.

MiMo-V2.5-Pro API pricing plans:

Note: MiMo also has a Token Plan that does not distinguish between 256K and 1M contexts and charges one flat rate, which suits users who need long context.

Step 2: Add MiMo in cc-switch

1. Double-click to open cc-switch

2. Under the Claude tab, click the plus sign in the top-right corner

3. For the provider, choose Xiaomi MiMo

4. Enter your API key

5. For the model name, enter mimo-v2.5-pro

Step 3: Enable and Start Using It

1. Once added, enable MiMo-V2.5-Pro on the cc-switch home screen
2. Open Claude Code and you can start using it immediately

Real-World Results

One user built a complete WeChat official account analytics platform with Claude Code + MiMo-V2.5-Pro:

1. Dictated the requirements by voice directly in Claude Code
2. The model produced a complete tech stack, architecture design, and data analysis logic
3. Automatically developed the frontend and backend and wired up a Feishu Base (multidimensional table) database
4. Finally deployed it to an internal server and set up enterprise permission management

Throughout the whole process, MiMo-V2.5-Pro performed close to what you would expect from Opus 4.6 -- organized output, smooth tool calling, and a deployment that passed on the first attempt.

Tip: if you have been coding with GLM-5.1 or Kimi K2.6, MiMo-V2.5-Pro is a new option worth trying alongside them. All three models have their own strengths, so switch flexibly depending on the specific task.

FAQ

• What is the difference between MiMo-V2.5 and MiMo-V2.5-Pro? The Pro version is the flagship model, with more parameters and stronger capabilities. The standard version is suited to lightweight tasks.

• How is the latency? Since it has just launched with few users, real-world latency is currently very low. It may change as the user base grows.

• Which programming languages are supported? All the mainstream ones. In real-world tests, web development (HTML/CSS/JS/React) was especially impressive.