Dear MiMo Team,

We are the team behind Agentbot — an open-source AI agent platform that deploys autonomous agents to Telegram, Discord, and WhatsApp. We process billions of tokens daily through MiMo-V2.5-Pro, and we wanted to share our experience and recommendations.

What MiMo Means to Us

MiMo-V2.5-Pro is the backbone of Agentbot. Every agent on our platform uses it for reasoning, code generation, and task execution. The model's efficiency — 3x reasoning improvement over previous versions — directly translates to lower costs for our users.

When we deployed MiMo, our token costs dropped dramatically. Users who were paying $50+/month for GPT-4 now pay $5 for the same work. That's not just a cost reduction — it's access democratization.

What's Working

Reasoning efficiency — 3x improvement is real. Our agents handle complex tasks faster.
Token pricing — The most competitive in the market. Our users love it.
OpenClaw compatibility — Seamless integration with the agent runtime.
Kingsoft Office — Document processing (Word, Excel, PPT, PDF) is a game-changer for agents.

Recommendations

Based on building Agentbot on MiMo, here are our recommendations for the next chapter:

1. Function Calling Improvements

MiMo excels at text generation but tool use could be more reliable. Better function calling accuracy would make it the default choice for agents that need to interact with external APIs.

2. Streaming Performance

For real-time agent UIs, streaming latency matters. The TileRT partnership showing 1000 tokens/s is incredible — bring that to the general API.

3. Multi-Modal Support

MiMo-V2-Omni is promising. For agents that need to understand screenshots, documents, or images, deeper vision integration would be valuable.

4. Agent-Specific Fine-Tuning

A model specifically fine-tuned for agent workflows — tool use, multi-step reasoning, error recovery — would be a killer feature. Agents are different from chatbots.

5. Better Documentation

The API docs are good but could use more agent-specific examples. Show how to build autonomous workflows, not just chatbots.

What We're Building Together

Agentbot is proof that MiMo works at scale. We have 120,000+ potential users, 5 integrated products, and a growing open-source community. Every agent on our platform runs on MiMo.

We're not just users — we're partners. And we want to help make MiMo the default model for autonomous agents worldwide.

With respect and gratitude,

The Agentbot Team

London, June 2026