MiMo-V2.5
MiMo-V2.5 is a large language model from Xiaomi, part of the company's MiMo family of models served through third-party inference APIs.
API Pricing
Cheapest on GMI Cloud — 52% below avg| Provider | Input / 1M | Output / 1M | Cached / 1M |
|---|---|---|---|
| $0.119 | $0.238 | $0.0024 | |
| $0.140 | $0.280 | $0.0028 | |
| $0.168 | $0.336 | - | |
| $0.168 | $0.336 | $0.0034 | |
| $0.199 | $0.638 | $0.100 | |
| $0.700 | $1.40 | $0.014 |
Prices updated daily. Last check: Oct 6, 2026
MiMo-V2.5 pricing by provider
Input, output, and batch rates, plus alternatives, for one provider at a time.
Performance & Benchmarks
Source: Artificial Analysis →Reasoning & Knowledge
- GPQA Diamond84.9%
- Humanity's Last Exam27.2%
Coding
- SciCode43.9%
Agentic & Tool Use
- Terminal-Bench Hard41.7%
- Terminal-Bench v2.163.7%
- τ²-bench90.6%
- τ-bench Banking8.7%
Instruction & Long Context
- IFBench67.1%
- Long-Context Reasoning73.0%
Benchmarks measured Oct 2026. Scores are independent evaluations, not vendor-reported.
Model Details
General
- Creator
- Xiaomi
- Modalities
- Text
Capabilities
- Tool Calling
- No
- Open Source
- Yes
Strengths & Limitations
Strengths
- Measured output speed of approximately 54.7 tokens per second in independent Artificial Analysis testing, adequate for streaming chat-style responses
- Adds a non-traditional vendor option to the hosted-model landscape, since Xiaomi's MiMo family sits outside the usual set of AI-lab model providers
- Served through third-party inference APIs, so pricing can be compared across providers rather than being fixed by a single vendor
- Part of a versioned family (MiMo-V2.5), making it straightforward to track against earlier and later MiMo releases
- Third-party throughput and latency measurements are available, which is not the case for every model in this size class
- Provider-level pricing for this model is listed side by side in the table on this page, simplifying cost comparison
Limitations
- Measured time to first token of roughly 2,315 ms is high, which is noticeable in latency-sensitive interactive or voice applications
- We do not track a confirmed context window for MiMo-V2.5, so long-document workloads need verification with the serving provider
- No standard reasoning, coding, or multilingual benchmark scores are recorded in our data, making capability comparison against peers difficult
- Modality support is not confirmed in our metadata — assume nothing about image or audio input without checking provider documentation
- Provider availability for MiMo models is narrower than for widely hosted open-weight families, which limits redundancy and price competition
Key Features
About MiMo-V2.5
Common Use Cases
MiMo-V2.5 is best evaluated for batch and asynchronous text workloads where a two-second wait before the first token is acceptable — content drafting, summarization, data extraction from moderate-length inputs, and offline classification or tagging jobs. Its measured throughput of roughly 54.7 tokens per second is workable for streamed chat responses but the long time to first token makes it a weaker fit for latency-critical paths such as voice agents, autocomplete, or multi-step agent loops where each hop pays that startup cost. Because we do not have confirmed context-window or benchmark data for MiMo-V2.5, teams considering it for coding, math, or long-context retrieval work should run their own evaluation against alternatives listed at similar prices in the table above rather than relying on published capability claims.
Frequently Asked Questions
How much does MiMo-V2.5 cost to use?
Pricing for MiMo-V2.5 varies by inference provider and by pricing type — input tokens, output tokens, and any cached or batch rates are typically billed separately. Because providers change rates and add or drop models frequently, check the pricing table on this page for current per-provider figures.
What is MiMo-V2.5 best used for?
It suits general text generation work where startup latency is not critical: drafting and rewriting content, summarization, structured extraction, and batch classification. Its measured first-token latency of around 2.3 seconds makes it less suitable for voice interfaces, autocomplete, or agent chains with many sequential model calls.
Who makes MiMo-V2.5?
MiMo-V2.5 is developed by Xiaomi as part of its MiMo model family. It is accessed through third-party inference providers, which are the entities listed in the pricing table on this page.
How fast is MiMo-V2.5?
Independent testing by Artificial Analysis measured roughly 54.7 output tokens per second with a time to first token of about 2,315 ms. Both figures depend on the serving provider, prompt length, and current load, so your observed numbers may differ.
What context window does MiMo-V2.5 support?
We do not currently have a confirmed context window for MiMo-V2.5 in our database. If your workload involves long documents or large codebases, verify the supported context length directly with the provider you plan to use.
How does MiMo-V2.5 compare to other models on this site?
Because we do not track published benchmark scores for MiMo-V2.5, the practical comparison points are its measured speed characteristics and its per-provider price relative to other models in the table above. For capability comparisons, running your own evaluation on representative prompts is the most reliable approach.