工具实测 | 小米MiMo-V2.5-Pro,降价后真香吗?Tool Review | Is the Xiaomi MiMo-V2.5-Pro Really a Bargain After the Price Drop?


AI free token/AI免费积分

小米MiMo-V2.5系列宣布永久降价,最高降幅99%,且不再区分输入长度。

降价后价格和DeepSeek基本一致,小米MiMo真的香吗?

接下来我们进入实测环节,模型使用MiMo-V2.5-Pro。

 

辅助编程测试

“生成一个旅游推荐网站首页图,可以按照时间和城市生成不同的推荐路线和攻略”

深度思考5.6秒,预览画面如下:

图片

推荐路线按照四季及热门旅游城市进行了划分,如果没有心仪目的地,还贴心增加了定制入口。

图片

 

动态效果测试

“做一个黑色的圆球,跟着鼠标运动轨迹进行运动”

深度思考2.5秒,效果如下:

工具实测 | 小米MiMo-V2.5-Pro,降价后真香吗?
已关注
关注
重播 分享 赞
 
拓扑森林OPC实战营
 
 
 

 

 

0/0

00:00/00:04
 
 
进度条,百分之0
 
 
00:00
/
00:04
00:04
倍速
全屏

 

倍速播放中
 
 

继续观看

工具实测 | 小米MiMo-V2.5-Pro,降价后真香吗?

 
转载
,
工具实测 | 小米MiMo-V2.5-Pro,降价后真香吗?
拓扑森林OPC实战营
已同步到看一看写下你的评论
 

 

 
 
视频详情

呈现效果还不错,基本能做到实时跟随鼠标轨迹。

 

逻辑推理测试

将一道初中数学题,以图片的形式上传,让MiMo解答。

深度思考26.8秒,给出答案。

图片

首先是多模态识图能力,完整理解题目,不只是文字。

解答思路正确,答案也对得上。

只有一个小问题,解题过程稍微麻烦了些。

虽然给出了正确答案,但不是最优解。

 

由于以上测试,思考时间都是秒级,大概率缓存命中。

为了得到更加真实的结果,进行复杂任务测试。

 

复杂任务测试

“设计一个动物中国地图,这个地图按照中国省份划分,每个省份都有代表动物”

第一次尝试提醒服务器繁忙,重新生成。

深度思考946.5秒,效果如下:

图片

34个省级行政区没有少,每个省也有代表动物,基本完成任务。

只是想再把画面优化更美观些,一直没有成功。

 

最后总结

通过测试不难发现,MiMo中文语境、文化适配高,推理速度并没有宣传那么快。

长链推理 “自我纠错” 不足,隐秘Bug排查、无明确提示时自主思考弱,需用户主动引导调整。

实际使用中的成本存在差异。

  • 缓存命中:若用户频繁调用相同或相似内容(如多轮对话、重复任务),缓存命中率高,实际成本极低,确实能大幅省钱。

  • 低缓存命中率场景:若每次调用都是全新输入(如处理大量不同文档),未命中缓存的单价较高,成本高于预期。

  • 输出密集型任务:输出价格虽降幅较大,但仍是成本较高的部分,若任务以生成大量文本为主,需关注输出Token消耗。

 

所以小米Mimo降价后,在大多数常见使用场景下(尤其是缓存命中率高的场景)确实能显著省钱。

但对于输出密集型或低缓存命中率的特殊场景,需根据实际Token消耗计算成本,没有想象中那么便宜。

建议根据自身使用模式,结合套餐额度和实际调用情况,综合评估成本效益。

 

如果这篇文章对你有帮助,记得点个"赞",分享给身边想用AI提效的朋友。

 

The Xiaomi MiMo-V2.5 series announces a permanent price cut, with the maximum discount reaching 99%, and no longer distinguishes between input lengths.

After the price drop, the price is almost the same as DeepSeek. Is Xiaomi MiMo really good value for money?

Next, we will enter the practical testing phase, using the model MiMo-V2.5-Pro.

Auxiliary programming test

"Generate an image for the homepage of a travel recommendation website, which can generate different recommended routes and strategies based on time and city."

Deep thinking for 5.6 seconds, the preview image is as follows:

图片

The recommended routes are divided by season and popular tourist cities. If you don't have a destination in mind, a customization entry is also thoughtfully added.

图片

Dynamic effect test

"Create a black ball that moves along the trajectory of the mouse cursor."

Deep thinking for 2.5 seconds, the effect is as follows:

Already followed

Follow

Replay Share Like

Close

Watch more

More

Exit fullscreen

Topology Forest OPC Practical Training Camp Already followed

Share video

Duration 00:04

0/0

00:00/00:04

Switch to horizontal screen mode

Continue playing

Progress bar, 0%

Play

00:00

00:04

00:04

Speed

Fullscreen

Speed playing

0.5x 0.75x 1.0x 1.5x 2.0x

Super clear Smooth

Continue watching

Tool review | Xiaomi MiMo-V2.5-Pro, is it really worth it after the price drop?

Watch more

Repost

,

Tool review | Xiaomi MiMo-V2.5-Pro, is it really worth it after the price drop?

Topology Forest OPC Practical Training Camp Already followed

Share Like Look

Already synchronized to Look Write your comment

Video details

The effect is not bad, basically able to follow the mouse trajectory in real-time.

Logical reasoning test

Upload a middle school math problem as an image, and let MiMo solve it.

Deep thinking for 26.8 seconds, providing the answer.

图片

First, the multimodal image recognition capability is strong, fully understanding the problem, not just the text.

The reasoning process is correct, and the answer matches.

There is only one small issue: the problem-solving process is a bit cumbersome.

Although the correct answer is given, it is not the optimal solution.

Since the thinking time for the tests above is in seconds, it is highly probable that the cache was hit.

To get more realistic results, perform a complex task test.

Complex task test

"Design an animal map of China. The map is divided by Chinese provinces, and each province has a representative animal."

The first attempt prompted a busy server; regenerated the image.

Deep thinking for 946.5 seconds, the effect is as follows:

图片

The map includes all 34 provincial administrative regions without missing any, and each province has a representative animal, basically completing the task.

However, trying to optimize the image quality failed repeatedly.

Final summary

From the tests, it is not difficult to see that MiMo has a high degree of adaptation to the Chinese context and culture, but its reasoning speed is not as fast as advertised.

Long-chain reasoning and "self-correction" are insufficient. In the absence of clear prompts, the ability to autonomously troubleshoot hidden bugs and think critically is weak, requiring active user guidance and adjustment.

There is a discrepancy in actual usage costs.

Cache hit: If users frequently call the same or similar content (such as multi-round dialogue or repeated tasks), the high cache hit rate results in extremely low actual costs, which can indeed significantly save money.

Low cache hit rate scenario: If each call is a completely new input (such as processing different documents), the cost for non-cached requests is high, exceeding expectations.

Output-intensive tasks: Although the output price has dropped significantly, it is still a high-cost part. If the task involves generating a large amount of text, attention must be paid to the consumption of output tokens.

Therefore, after the price cut, Xiaomi Mimo is indeed more cost-effective in most common usage scenarios, especially those with high cache hit rates.

However, for output-intensive or low-cache-hit-rate scenarios, the cost depends on the actual token consumption, which may be higher than anticipated.

It is recommended to evaluate the cost-effectiveness based on the specific usage mode, package allowance, and actual invocation frequency.

If this article is helpful, please give it a "like" and share it with friends who are looking to improve efficiency with AI.

Tool Review | Is the Xiaomi MiMo-V2.5-Pro Really a Bargain After the Price Drop?

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry