April 23rd.MilletToday, the Xiaomi MiMo-V2.5 series is officially launchedPublic BetaI don't know. The MiMo-V2.5-Pro and MiMo-V2.5 models are about to open the world。
1AI was informed from the bulletin that the Xiaomi MiMo-V2.5 series consists of MiMo-V2.5, V2.5-Pro, V2.5-TTS Series, V2.5-ASR, with a stronger reasoning, a more stable Agent, a longer context, stronger commands that are understood by vague commands, and a better general sense and understanding. At the same time, Mi also optimized the Token Plan pricing programme。
MiMo-V2.5-Pro: Stronger intelligence, longer focus
MiMo-V2.5-Pro was described as the most powerful model of millet to date. YesUniversal intelligence capabilities, complex software engineering and long-range missionsIn equal dimensions, it has been able to make a full leap forward from the top global Agent model (Claude Opus 4.6, GPT-5.4)。
In-house testing, MiMo-V2.5-Pro is fitted with a suitable operating framework that stabilizes long-range missions involving a single round of nearly 1,000 tools, and command compliance in an intelligent body scene is significantly enhanced - both to capture the hidden requirements in the context and to maintain logical consistency over a long period of time. As a result, MiMo-V2.5-Pro has been able to take on a truly serious professional job with greater confidence。
MiMo-V2.5-Pro was created for more difficult and complex mission objectives. Mi entrusted it with tasks that would take days or even weeks for human experts to complete a long, independent journey and remain of high quality. The following results were delivered:
- Full Sysy compiler with Rust
This task, which originated from the Beijing University ' s " Principles of Compiler " course project, requires that the model achieves a complete SysY compiler with Rust from zero: syntax analyser, syntax analyser, AST, Koopa IR code generation, RISSC-V compilation backend, and performance optimization. As a referenceIt's usually a few weeks before the North Big undergraduates finish the project MiMo-V2.5-Pro Time only 4.3 Hours, 672 times through the tool to complete all work and get on the hidden test set The full score of 233/233 shows an extremely efficient productivity value。
It is not caught in the force of repeated error, but it builds the entire compiler on a layer by layer: first complete flow line skeleton, then layer by layer - Koopa IR (110/110), RISC-V back end full (103/103), performance optimizes full score (20/20). It was first compiled and passed 137/23359%'s cold-start pass rate means that the structure is right before running any tests. Round 512, a reorganisation order Iv9/riscv returned two test points; the model is self-diagnosed, restored and continues to advance。
- Develop a video editor
Only a few simple commands — “Construct a video editor Web application — MiMo-V2.5-Pro delivers a runable Web application with multi-orbit time lines, clippings, cross-declining, audio mixing and export processes. The number of codes that are eventually constructed is 8,192 rows, which have been called 1,868 times by tools and are completed in 11.5 hours of autonomous work。
MiMo-V2.5: Overstep full-modular Agent, millions of context
MiMo-V2.5 is a whole-of-life model for the Agent sceneLarge ModelThey can read, hear and read at the same time and translate understanding into action. This time, MiMo-V2.5 brings a key upgrade:
Agent Capability fully exceeds MiMo-V2-Pro
In authoritative Agent assessments such as Claw-Eval, MiMo-V2.5 exceeds MiMo-V2-Pro to perform simple daily tasks, while API costs are reduced by approximately 50%。
Multimodular senses are all over MiMo-V2-Omni
Increased capacity in cross-modular reasoning, video understanding, graphic analysis, etc., approaches and even exceeds industry top-level closed-source models in VideoMME, CharXiv, MMMU-Pro, etc。
MiMo-V2.5 Full Series: More Token Efficiency
The entire MiMo-V2.5 series optimizes token efficiency and does more with fewer Token。
With the same Agent benchmark ClawEval score:
- MiMo-V2.5-Pro savings from Kimi K2.6
- MiMo-V2.5 Savings from Muse Spark 50% Token
MiMo-V2.5 Full series: How to use it
- MiMo-V2.5-Pro was created for the Agent mission, MiMo-V2.5 covers most generic Agent scenes
- MiMo-V2.5 Supports original whole-state Agent capability, covering images, audio and video
- MiMo-V2.5 with a higher average reasoning speed to respond more quickly to time-sensitive tasks
Token Plan Upgrade
We've made a few substantive improvements for Token Plan:
Credits Rate update, more favourable
- MiMo-V2.5:1x (consumption 1 Token = 1 Credit)
- MiMo-V2.5-Pro: 2x (consumption 1 Token = 2 Credits)
Cancel 1 Token = 4 Credits billing, nowToken Plan No more distinction 256k and 1 M Context Window Credit Triple rate。
Optimal preferential rate at night
Beijing time at 00:00 ~ 08:00, all models Credits consumption rateFight again on the basis of the original 8 Debits.
Auto-renewal discount
- Add a new "suspensive month" subscription mode, with the next month 7 discounts for old users and 77 discounts for new users, with an average of one。
- Adds a new "package year" subscription cycle, with one subscription with 88 discounts for the entire year, without supersing the initial purchase/ automatic renewal fee。
Online benefits: Token Plan user Credits fully reset
All purchased Token Plan users (as of April 22nd, Beijing time) The Credits level will be replaced by zero, recalculate。