Xiaomi MiMo-V2-Pro: A Lower-Cost Frontier Model With Strong Agentic Performance
Xiaomi announced MiMo-V2-Pro, described as a 1T-parameter sparse model (42B active per forward pass) with a 1M-token context window. Public reporting and cited benchmark snapshots position it as competitive in agentic and coding tasks, with lower listed API pricing than several premium frontier models.
What is reported so far
- Architecture: sparse MoE-style setup with 1T total parameters and 42B active parameters.
- Context: up to 1M tokens (with pricing tiers split at 256K context).
- Benchmark positioning: reported strong results on agent-focused benchmarks; benchmark outcomes remain workload-dependent.
- Third-party references: Artificial Analysis rankings were cited in launch coverage.
Pricing details (as reported)
- MiMo-V2-Pro (≤256K context): $1.00 input / $3.00 output per 1M tokens
- MiMo-V2-Pro (256K–1M context): $2.00 input / $6.00 output per 1M tokens
- GPT-5.2 (comparison table): $1.75 input / $14.00 output per 1M tokens
- Claude Opus 4.6 (comparison table): $5.00 input / $25.00 output per 1M tokens
Interpretation for teams
The main signal is cost-capability positioning: lower token pricing with reported near-frontier benchmark performance in selected categories. This is a practical reason to include MiMo-V2-Pro in controlled evaluations, using your own prompts, safety requirements, and latency targets before production decisions.