Summary
MiniMax has released M2.5, its new flagship coding model, positioned as a benchmark target against Anthropic's Claude Opus 4.6. According to the announcement (sourced from IT之家), M2.5 is described as the world's first production-grade model designed for agent scenarios, with full-stack support spanning PC, app, and cross-platform development, plus industry-leading performance in office tasks such as Excel, PowerPoint, and research. The model uses 10 billion activated parameters for efficient, lightweight operation and delivers throughput up to 100 tokens per second for fast inference. Following the launch, MiniMax's stock surged more than 20%, pushing its market capitalization past HKD 180 billion, reflecting strong market recognition. The poster-style announcement was published on a Chinese tech forum with a February 12, 2026 date, highlighting MiniMax's push into agentic and developer-focused AI tooling.
MiniMax M2.5 Flagship Coding Model Launch
A Chinese tech forum post (sourced from IT之家, dated 2026.02.12) announces the release of MiniMax M2.5, MiniMax AI's new flagship programming model.
Key Points
- Benchmark target: The announcement positions M2.5 against Claude Opus 4.6 as its reference benchmark.
- Agent-native design: Marketed as the world's first production-grade model designed specifically for agent scenarios.
- Full-stack support: Claims all-round development capability across PC, app, and cross-platform targets.
- Office SOTA: Industry-leading performance on office/productivity tasks including Excel, PowerPoint, and research.
Technical Specs (as stated)
| Metric | Value |
|---|---|
| Activated parameters | 10B (efficient, lightweight) |
| Throughput | 100 TPS (ultra-fast inference) |
Market Reaction
- MiniMax's stock rose more than 20% following the launch.
- Market capitalization surpassed HKD 180 billion, described as strong market recognition.
*Source: IT之家, as cited in the original forum post. Figures above reflect the poster's claims.*
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/176922705