Published February 13, 2026

MiniMax M2.5: Open-Source Models Rival Claude Sonnet

MiniMax M2.5: Open-Source Models Catch Up to Claude Sonnet

Chinese company MiniMax has released M2.5, a family of open-weight models whose performance is approaching that of Claude 3.5 Sonnet.

Products
Event Source: OpenHands Reading Time: 3 – 5 minutes

Chinese company MiniMax has released the M2.5 family of models – one of those rare cases where open-source models genuinely approach the quality of proprietary solutions like Claude 3.5 Sonnet. Simply put, the gap between models you can run on your own hardware and those available only through major companies' APIs is gradually narrowing.

Что такое модели MiniMax M2.5 и их особенности

What is MiniMax M2.5?

MiniMax is a Chinese company focused on developing large language models. The new M2.5 family includes several versions of different sizes, from compact 7-billion-parameter models to large 671-billion-parameter ones. They are all released with open weights, meaning any developer or researcher can download and use them locally.

The main highlight of this release is its performance. According to the developers' internal tests, the largest model in the family shows results comparable to Anthropic's Claude 3.5 Sonnet. This is significant because, until now, open-source models have typically lagged noticeably behind closed-source solutions.

Как тестировались возможности MiniMax M2.5

How Was It Tested?

The OpenHands team – a project developing autonomous AI agents for programming – tested MiniMax M2.5 on their own tasks. OpenHands uses language models to automate routine developer tasks like fixing bugs, writing code, and working with repositories.

In these tests, the MiniMax M2.5 model achieved results close to Claude 3.5 Sonnet on programming-related tasks and in solving real-world coding problems. These aren't abstract benchmarks but practical scenarios encountered in day-to-day work.

It's important to understand this doesn't mean MiniMax has completely surpassed all its competitors. However, it has genuinely reached a level where an open-source model can be a viable alternative to commercial solutions for specific tasks.

Значение MiniMax M2.5 для развития открытых моделей ИИ

Why Is This Significant?

Open weights are not just a matter of ideology or principle. They provide the ability to run a model on your own infrastructure, control your data, adapt it to your specific needs, and avoid dependency on the availability of third-party APIs.

Until recently, this freedom came with a trade-off: open-source models performed noticeably worse than closed-source ones. The difference was so significant that, for many tasks, choosing API-based solutions was the obvious choice, despite all their limitations.

MiniMax M2.5 shows that this trade-off is becoming less painful. If an open-source model can produce results close to Claude 3.5 Sonnet, it changes the game for developers building AI-based products.

Различные версии MiniMax M2.5 для разных задач

Multiple Versions for Different Tasks

The M2.5 family includes models of various sizes. The smallest, at 7 billion parameters, is suitable for quick tasks and can even run on consumer hardware. The mid-sized versions offer a compromise between speed and quality. The largest model, at 671 billion parameters, requires significant resources but also delivers the best performance.

This approach allows users to choose the right model for a specific task. If you need high speed and can accept slightly lower accuracy, you can use a smaller version. If performance is critical and you have the resources, use the largest one.

Будущее открытых моделей и MiniMax M2.5

What's Next?

The release of MiniMax M2.5 is another step toward parity between open-source and closed-source models. We're not at full equality yet: proprietary solutions still lead in a number of tasks, especially where stability, security, and fine-tuning model behavior are crucial.

But the gap is closing. And that's good news for everyone building AI-based products: the choices are expanding, and the dependence on specific vendors is shrinking.

It's too early to say if MiniMax M2.5 will become the new standard for open-source models. But the fact that such solutions are emerging and demonstrating competitive results is important in itself. It means the market is moving toward greater openness – and this movement is gaining momentum.

Original Title: MiniMax M2.5: Open Weights Models Catch Up to Claude Sonnet
Publication Date: Feb 12, 2026
OpenHands openhands.dev An open-source project developing AI agents for software engineering and automation tasks.
Previous Article LightOn Unveils Code Search Tool That Understands Queries Semantically Next Article Training Language Models with Feedback: verl Now Runs on AMD GPUs

From Source to Analysis

How This Text Was Created

This material is not a direct retelling of the original publication. First, the news item itself was selected as an event important for understanding AI development. Then a processing framework was set: what needs clarification, what context to add, and where to place emphasis. This allowed us to turn a single announcement or update into a coherent and meaningful analysis.

Neural Networks Involved in the Process

We openly show which models were used at different stages of processing. Each performed its own role — analyzing the source, rewriting, fact-checking, and visual interpretation. This approach maintains transparency and clearly demonstrates how technologies participated in creating the material.

1.
Claude Sonnet 4.5 Anthropic Analyzing the Original Publication and Writing the Text The neural network studies the original material and generates a coherent text

1. Analyzing the Original Publication and Writing the Text

The neural network studies the original material and generates a coherent text

Claude Sonnet 4.5 Anthropic
2.
Gemini 2.5 Pro Google DeepMind step.translate-en.title

2. step.translate-en.title

Gemini 2.5 Pro Google DeepMind
3.
Gemini 2.5 Flash Google DeepMind Text Review and Editing Correction of errors, inaccuracies, and ambiguous phrasing

3. Text Review and Editing

Correction of errors, inaccuracies, and ambiguous phrasing

Gemini 2.5 Flash Google DeepMind
4.
DeepSeek-V3.2 DeepSeek Preparing the Illustration Description Generating a textual prompt for the visual model

4. Preparing the Illustration Description

Generating a textual prompt for the visual model

DeepSeek-V3.2 DeepSeek
5.
FLUX.2 Pro Black Forest Labs Creating the Illustration Generating an image based on the prepared prompt

5. Creating the Illustration

Generating an image based on the prepared prompt

FLUX.2 Pro Black Forest Labs

Related Publications

You May Also Like

Explore Other Events

Events are only part of the bigger picture. These materials help you see more broadly: the context, the consequences, and the ideas behind the news.

Want to know about new
experiments first?

Subscribe to our Telegram channel — we share all the latest
and exciting updates from NeuraBooks.

Subscribe