MiMo-V2.5-Pro

Xiaomi's open-weights flagship (April 2026): a 1.02T-parameter MoE (42B active, MIT license) built for long-horizon autonomous tasks with 1,000+ tool calls. Top-20 by OpenRouter usage and #5 open-weights on the Artificial Analysis Intelligence Index. Throughput is modest (~47 tok/s), but capability-per-dollar is exceptional.

MiMo-V2.5-Pro strengths

• MIT open weights (1.02T MoE)
• 1,000+ tool-call long-horizon agents
• Strong SWE-bench Pro results
• 1M context
• Exceptional capability-per-dollar

Pricing & context

Cost guide: a typical call of about 10K input + 2K output tokens costs roughly $0.006 at list prices. Worth modelling against cheaper tiers before committing high-volume traffic.

When to choose MiMo-V2.5-Pro

MiMo-V2.5-Pro is best for agentic coding and long-running autonomous agents on a budget, including self-hosted deployments. If your workload is more cost-sensitive, weigh it against gpt-oss-120b (≈$0.03 input /1M) first.

MiMo-V2.5-Pro FAQ

How much does MiMo-V2.5-Pro cost?

MiMo-V2.5-Pro is priced at $0.435 per 1M input tokens and $0.87 per 1M output tokens (public API list price), with a 1M tokens context window. A typical call of about 10K input and 2K output tokens costs roughly $0.006.

What is MiMo-V2.5-Pro best for?

MiMo-V2.5-Pro by Xiaomi is best for agentic coding and long-running autonomous agents on a budget, including self-hosted deployments.

How does MiMo-V2.5-Pro pricing compare to GPT-5.6 Luna?

MiMo-V2.5-Pro input costs $0.435 per 1M tokens versus $1.00 for GPT-5.6 Luna, roughly 2.3x less expensive on input. Output is $0.87 vs $6.00.

Is MiMo-V2.5-Pro multimodal?

MiMo-V2.5-Pro supports text.