MiniMax M2.5: New Open-Source AI Model Cuts Costs by 95% | VentureBeat

by Chief Editor

The AI Price War is Here: MiniMax’s M2.5 and the Democratization of Intelligence

The artificial intelligence landscape shifted dramatically this week with the release of MiniMax’s M2.5 language model. This isn’t just another AI; it’s a potential game-changer, promising to deliver performance rivaling industry leaders like Anthropic’s Claude Opus 4.6 at a fraction of the cost – up to 95% less, according to MiniMax.

From Chatbots to AI Workers: A Paradigm Shift

For years, accessing cutting-edge AI has felt like a luxury, constrained by high costs and token limits. MiniMax is betting on a different future: one where AI is so affordable it becomes ubiquitous, moving beyond simple question-and-answer interactions to develop into a true “worker” embedded in everyday tasks. What we have is already happening internally at MiniMax, where M2.5 now completes 30% of all tasks and generates 80% of newly committed code.

The Technology Behind the Breakthrough: Sparse Power and Forge

The secret to M2.5’s efficiency lies in its Mixture of Experts (MoE) architecture. Instead of activating all 230 billion parameters for every task, it selectively engages only 10 billion, balancing power and agility. This is coupled with a proprietary Reinforcement Learning (RL) framework called Forge, which allows the model to learn from simulated real-world environments. MiniMax engineer Olive Song highlighted the importance of Forge in scaling performance, even with a relatively slight number of parameters.

A key component of Forge is the CISPO (Clipping Importance Sampling Policy Optimization) mathematical approach, designed to stabilize training and foster an “Architect Mindset” – enabling M2.5 to proactively plan projects before diving into code.

Benchmarking the Performance: Competing with the Best

Early benchmarks show M2.5 holding its own against top-tier models. It achieves 80.2% on the SWE-Bench Verified test, matching Claude Opus 4.6’s speed. It also demonstrates leading performance in areas like BrowseComp (76.3%) and Multi-SWE-Bench (51.3%). The model’s efficiency is striking: tasks that cost $3.00 with Claude Opus 4.6 can be completed for around $0.15 with M2.5.

MiniMax M2.5 line plot comparing different models performance over time on SWE benchmark. Credit: MiniMax

Pricing and Accessibility: A New Era for AI Adoption

MiniMax offers two API versions: M2.5-Lightning (optimized for speed) and Standard M2.5 (optimized for cost). The pricing is significantly lower than competitors, with the potential to run four “agents” continuously for a year for around $10,000. Here’s a pricing comparison:

Model Input Output Total Cost Source
Qwen 3 Turbo $0.05 $0.20 $0.25 Alibaba Cloud
deepseek-chat (V3.2-Exp) $0.28 $0.42 $0.70 DeepSeek
deepseek-reasoner (V3.2-Exp) $0.28 $0.42 $0.70 DeepSeek
Grok 4.1 Fast (reasoning) $0.20 $0.50 $0.70 xAI
Grok 4.1 Fast (non-reasoning) $0.20 $0.50 $0.70 xAI
MiniMax M2.5 $0.15 $1.20 $1.35 MiniMax
MiniMax M2.5-Lightning $0.30 $2.40 $2.70 MiniMax
Gemini 3 Flash Preview $0.50 $3.00 $3.50 Google
Kimi-k2.5 $0.60 $3.00 $3.60 Moonshot
GLM-5 $1.00 $3.20 $4.20 Z.ai
ERNIE 5.0 $0.85 $3.40 $4.25 Baidu
Claude Haiku 4.5 $1.00 $5.00 $6.00 Anthropic
Qwen3-Max (2026-01-23) $1.20 $6.00 $7.20 Alibaba Cloud
Gemini 3 Pro (≤200K) $2.00 $12.00 $14.00 Google
GPT-5.2 $1.75 $14.00 $15.75 OpenAI
Claude Sonnet 4.5 $3.00 $15.00 $18.00 Anthropic
Gemini 3 Pro (>200K) $4.00 $18.00 $22.00 Google
Claude Opus 4.6 $5.00 $25.00 $30.00 Anthropic
GPT-5.2 Pro $21.00 $168.00 $189.00 OpenAI

The Rise of Chinese AI: A New Global Force

MiniMax’s advancements signal a growing strength in the Chinese AI sector. Zhipu, another Chinese AI company, recently released its GLM-5 model and saw its stock surge 30% as a result. This suggests that Chinese companies are rapidly closing the gap with their US counterparts in AI development, despite having fewer resources in terms of GPUs.

What Does This Imply for the Future?

MiniMax’s M2.5 isn’t just a technological achievement; it’s a strategic shift. It lowers the barrier to entry for AI adoption, enabling businesses of all sizes to leverage the power of large language models. The focus is moving from building the most intelligent AI to building the most *useful* and *accessible* AI. This could unlock a wave of innovation, as developers are freed from the constraints of cost and can focus on creating truly transformative applications.

Frequently Asked Questions

  • What is Mixture of Experts (MoE)? MoE is an architecture where only a portion of the model’s parameters are activated for each task, improving efficiency.
  • What is Forge? Forge is MiniMax’s proprietary Reinforcement Learning framework used to train M2.5.
  • How does M2.5 compare to Claude Opus 4.6? M2.5 offers comparable performance to Claude Opus 4.6 at a significantly lower cost.
  • Is M2.5 truly open source? MiniMax claims it is, but the weights and license terms haven’t been released yet.

Pro Tip: Explore the MiniMax API documentation to understand how you can integrate M2.5 into your projects and take advantage of its cost-effectiveness.

What are your thoughts on the potential impact of MiniMax’s M2.5? Share your insights in the comments below!

You may also like

Leave a Comment