Xiaomi is getting ready to launch HyperOS 3, its new custom software based on Android 16. But before that, the company has decided to stop supporting a number of older smartphones. In its latest internal update, Xiaomi has officially added more models to its End-of-Life (EOL) list. This means these phones will no longer get software updates from the company.

These Xiaomi Phones Will No Longer Get Updates

As Google continues rolling out Android 16, Xiaomi is focusing on its own version of the software — HyperOS 3. Some of Xiaomi’s latest models like the Xiaomi 15 and Xiaomi 14T Pro have already received the Android 16 update through HyperOS 2.3. But not every phone will be that lucky.

Xiaomi has now confirmed that the following phones will not receive any future software updates:

• Xiaomi 11T Pro
• Xiaomi 11T
• Xiaomi 11 Lite LE
• Xiaomi 11 Lite 5G NE
• Redmi 11 Prime 4G
• Redmi A1+
• Redmi A1
• POCO M5
• POCO C50

What This Means for Users

These phones will still get security patches until September 2025, so they won’t be completely outdated right away. However, they won’t get any new features or major Android updates, including HyperOS 3. Some of these models didn’t even receive the earlier HyperOS 2 update.

As of now, Xiaomi hasn’t shared an official roadmap for the release of HyperOS 3 on supported devices. But it is expected to roll out in the coming months.

So if you’re using one of the phones on the list, it might be time to start thinking about an upgrade.

In a recent speed test comparison, the POCO F7 outperformed the OnePlus Nord 5 in multiple real-world performance checks. Both smartphones are priced similarly and come with 12GB RAM, but the POCO F7 clearly showed better optimization and faster results in most tasks.

Software and Specs

• POCO F7 runs on HyperOS 2.0 with the latest Snapdragon 8s Gen 4 processor.
• OnePlus Nord 5 features OxygenOS 15 and a Snapdragon 8s Gen 3 processor.
• Both phones were tested on equal settings—120Hz refresh rate, no virtual RAM advantage, and same Wi-Fi connection.

App Opening and RAM Management

POCO F7 performed better in app opening speed in most cases. It was faster in launching apps like Chrome, Instagram, Twitter, Amazon, and games like BGMI and Grid Autosport. However, OnePlus Nord 5 did better in opening the camera app and fingerprint unlock speed.

In RAM management, the POCO F7 again showed stronger performance. More apps and games stayed in memory without reloading, while the Nord 5 often reloaded them—indicating weaker RAM handling.

Storage Speed and Video Editing

A clear difference was observed in storage performance. The POCO F7, with its UFS 4.1 storage, extracted and copied large files much faster than the UFS 3.1 storage on the Nord 5. In a 4K video export test, both phones performed similarly, with POCO F7 finishing slightly faster.

Gaming Test

The gaming performance varied. While POCO F7 opened BGMI and Grid Autosport faster, OnePlus Nord 5 had a slight edge in loading Call of Duty. But overall, POCO F7 maintained more games in memory, which is better for multitasking gamers.

Final Thoughts

Overall, the POCO F7 offers better performance, faster app handling, and superior storage speed compared to the OnePlus Nord 5. OnePlus Nord 5 also caused some issues during testing by unexpectedly killing background apps, which interrupted the video shoot multiple times.

Still, performance can vary based on usage.

Xiaomi MiMo-7B marks a major leap for the Chinese tech giant into the world of artificial intelligence. Announced as Xiaomi’s first open-source large language model (LLM), MiMo-7B is engineered with a strong emphasis on reasoning and coding capabilities. While many LLMs compete for scale, Xiaomi takes a different approach — delivering competitive performance in a leaner, 7-billion parameter model that rivals even some heavyweight models like OpenAI’s o1-mini and Alibaba’s Qwen.

This bold move reflects Xiaomi’s growing commitment to innovation beyond smartphones and smart home devices, stepping firmly into the AI research and developer tooling space.

Optimized Pre-Training for Smarter Reasoning

At the core of MiMo-7B’s capabilities is a highly optimized pre-training process, which involved 25 trillion tokens — an unusually large training set for a model of this size. Of these, a dedicated 200 billion tokens focused solely on reasoning tasks. This data-centric approach laid a strong foundation for the model to understand logic, structure, and context effectively.

One standout aspect of the training method is Xiaomi’s use of Multiple Token Prediction (MTP), where the model learns to predict more than one token at a time. This unique mechanism enhances inference speed without degrading the quality of responses, especially in longer conversational or coding contexts.

Training Infrastructure and Reinforcement Learning

To push MiMo-7B to the next level, Xiaomi employed reinforcement learning (RL) techniques post-pretraining. These techniques were fine-tuned using a novel Test Difficulty Driven Reward system, which adapts based on the complexity of the task. This allows the model to get more nuanced training feedback, helping it improve in real-world reasoning scenarios.

The model also benefits from Xiaomi’s Easy Data Re-Sampling strategy, which dynamically revisits difficult data points to reinforce stability and learning.

On the infrastructure side, Xiaomi developed a proprietary Seamless Rollout Engine — a backend training optimization tool that reduces GPU idle time. As a result, MiMo-7B’s training speed improved by 2.29x, and its validation performance nearly doubled compared to conventional methods.

MiMo-7B Benchmark Performance

Xiaomi MiMo-7B has already been benchmarked across multiple high-level tasks, with promising results:

• Mathematics: 95.8% on the MATH-500 benchmark and 68.2% on the 2024 AIME dataset — both above-average scores for models in its class.

• Coding: Achieved 57.8% on LiveCodeBench v5 and 49.3% on v6, confirming its strength in code generation and problem-solving.

• General Knowledge & Reasoning: Solid mid-to-high 50s across benchmarks like DROP, MMLU-Pro, and GPQA.

These scores position MiMo-7B as a serious player in compact LLMs optimized for developers, researchers, and educational applications.

Also Read: Apple M4 MacBook Air Hits All-Time Low Price: Unprecedented Deal for 2025

Four Variants for Flexibility

To make MiMo-7B more accessible and usable across various use cases, Xiaomi has released four different model variants:

1. Base – The raw, pre-trained version of MiMo-7B.

2. SFT – Supervised fine-tuned version, ideal for controlled response generation.

3. RL-Zero – Reinforcement learning applied directly to the base model.

4. RL – Fine-tuned + reinforcement-learned, providing the best accuracy and stability for complex use cases.

All four variants are available under an open-source license, encouraging collaboration and transparency. This step also invites independent developers and research institutions to contribute, test, and even deploy these models in their own applications.

Conclusion

The release of Xiaomi MiMo-7B shows that small doesn’t mean underpowered. With thoughtful data curation, efficient training methods, and benchmarked performance, Xiaomi has crafted a 7B parameter model that holds its own in the world of AI — especially in reasoning-heavy and coding-focused tasks.

By making it open-source, Xiaomi is also signaling its commitment to the global AI developer ecosystem. MiMo-7B isn’t just a product — it’s a platform for innovation, collaboration, and community-driven progress.

For developers, researchers, and enterprises looking for an efficient yet powerful AI model, Xiaomi MiMo-7B is worth serious consideration.