Xiaomi MiMo V2.5 Pro

Xiaomi MiMo V2.5 Pro is Xiaomi’s flagship model with 1.02 trillion total and 42 billion active parameters, designed for frontier reasoning and agentic workflows. Its hybrid attention architecture significantly reduces KV-cache memory, and the context window spans one million tokens. Natively omnimodal for text, image, video, and audio, and fully commercially usable under the MIT license.

Xiaomi Version V2.5-Pro Commercial use permitted MoE 1020 B (42 B active) 1024 K Context 05/2025 $0.435 / $0.87 per 1M

Open Weights
Frontier
OpenRouter
Text
Vision
Video
Audio
Instruction-Tuned
Agentic Orchestrator
Interactive

Sovereign Risk: MEDIUM Xiaomi is a Chinese company and subject to China’s Data Security Law (DSL) and National Intelligence Law (NIL). The weights are publicly available under the MIT license. When using cloud services, state access to transmitted data is theoretically possible. Local deployment with the public weights reduces the risk — the NIL is only directly relevant when using a cloud API.

Tool-use details

Asset performance, reliability, and runtime profile

Name	P1	P2	Combined
EU License Research	100	15	49
HTTP Fetch & Extract	80	80	80
Tool Failure Handling (404)	80	60	73
Web Search & Tool Selection	100	60	82
URL Construction & Fetch	80	80	80
Multilingual Search & Synthesis	100	60	80

Name

Combined

EU License Research

100

HTTP Fetch & Extract

Tool Failure Handling (404)

Web Search & Tool Selection

100

URL Construction & Fetch

Multilingual Search & Synthesis

100

CrucibleMark evaluates tool use across 6 independent tests. Click a test name for details.

Reliability

Tool Call Valid: No

Retry: Not required

Hallucination: Detected

Reliability measures how consistently a model actually executes tool calls: Tool Call Valid schema and format accepted, Retry Required successful only after retry, Hallucination Flag fabricated tools or parameters detected. All three green means production-ready.

Runtime profile

Call 1: 7.01; First Request

MCP: 1.11; Protocol Latency

Synthesis: 23.68; Response Generation

Total: 190.8; Sum of All Phases

Token: 16076; Input + Output

Cost: $0; Cost per Run

The runtime profile shows the latency and cost metrics for the model run: Call 1 First Request, MCP Protocol Latency, Synthesis Response Generation, Total sum of all phases, supplemented by Token input and output and Cost cost per run.

Deployment Verdict

Conditional deploy: tool execution is strong, but a hallucination on the honeypot and one invalid tool call limit confidence for production MCP pipelines. The combined score of 74 shows usable substance, but falls short of unsupervised high-trust workflows.

Tool Execution Profile

Xiaomi MiMo V2.5 Pro demonstrates genuine tool intelligence. On the Web Search & Tool Selection test — which checks whether the model selects web_search over fetch without being prompted — it reliably identifies the correct path. This argues against rigid schema behavior. It also consistently invokes the required tools on Multilingual Search & Synthesis and EU License Research.

Protocol adherence is less clean. The global finding “tool call valid: false” is relevant for MCP integration, since even minor format or parameter errors cause orchestrators to fail hard. The absence of a required retry speaks less to a simple formatting slip and more to localized imprecision in execution. On the URL Construction test — which measures correct URL derivation followed by a fetch — it performs adequately, but not deterministically enough for fragile pipelines.

Synthesis Fidelity

How well does it consolidate tool results? Only to a limited degree. The P2 score of 44.17 is this model’s clear weak point. Strong individual results on HTTP Fetch & Extract and URL Construction & Fetch show it can cleanly consolidate structured content from real tool responses. As soon as the task demands more evaluation, compression, or multi-step synthesis, reliability visibly drops.

Does it stay within tool results or fall back on training data? Not consistently. On the honeypot EU License Research — which tests whether current license restrictions are answered from web sources rather than from the training snapshot — the model fails clearly with P2=15 and a confirmed hallucination. This is not merely a quality shortcoming; it is a security risk. When a model presents fabricated or pre-trained facts as the output of a live search, it corrupts the chain of evidence for the entire tool infrastructure.

Error Resilience

On the 404 test — which measures transparent behavior when a tool call fails — the model responds acceptably. It does not hallucinate page content despite the error. P2=60 is not elegant, but production-viable, because the fundamental rule is observed: surface the error rather than generate substitute facts.

Operational Profile

Call 1: 7.01s. MCP latency: 1.11s. Call 2: 23.68s. Total: 190.80s. Slow for the level of synthesis quality shown. Cost/run: local. Model price: $0.435 per 1M input tokens, $0.87 per 1M output tokens. Economically defensible as an Open Weights model, but the runtime is high for interactive orchestration.

Conclusion & Recommendation

Suitable for supervised agent pipelines where tool selection, web research, and multilingual retrieval matter more than reliable end-stage synthesis. Not suitable for compliance, licensing, policy, or other evidentiary pipelines where the model must remain strictly bound to tool results. If you deploy it, do so with hard output gating, tool-result quoting, and downstream verification before any externally consequential decision.

This analysis was generated automatically based on benchmark data. Model used: GPT 4.5 by OpenAI. The raw data and full methodology are documented in the GitHub project.