MiniMax M2.5 arrives on the market at a particularly delicate historical moment for the artificial intelligence industry. Initial euphoria for large language models certainly persists, but growing in parallel is also an increasingly widespread skepticism about astronomical costs and concrete, actually measurable results.

Codemotion's platform is a central hub for developers, offering insights into software development, technology trends, and coding tutorials. Through articles, tutorials, and conferences, Codemotion offers insights into programming languages, development frameworks, and best practices in software engineering. Developers can learn about software design principles, development methodologies, and emerging technologies to enhance their coding skills and build high-quality software applications.

Codemotion

MiniMax M2.5, launched February 12, 2026, is a 230B parameter Mixture-of-Experts model that activates only 10B parameters per query, achieving near-parity with Claude Opus 4.6 and GPT-5 on coding benchmarks like SWE-Bench Verified (80.2%) at roughly 1/20th the output cost ($1.20 vs $25 per million tokens). The Lightning variant runs at 100 tokens/second for $0.30 input/$2.40 output per million tokens. MiniMax completed a $619M Hong Kong IPO with shares doubling on day one to a ~$13B market cap. The model features an 'architect mindset' trained across 200,000+ real environments, supports 10+ programming languages, and handles 204,800-token context windows. Full weights are open-sourced on Hugging Face under a modified MIT license. The piece contextualizes M2.5 within the broader pattern of Chinese AI labs turning US chip embargo constraints into efficiency-driven innovation advantages, while raising questions about data sovereignty, long-term financial sustainability, and regulatory fit under the EU AI Act.

MiniMax M2.5: low costs, high performance, relaunches the Chinese AI geopolitical challenge

The Year of Maturity: AI in 2026 Between Autonomous Agents, Sovereignty, and the Reinvention of Work