PulseAugur
LIVE 13:49:46
research · [1 source] ·
0
research

MiniMax M2.5 open-sources agent-native RL model, rivaling Sonnet on coding

By PulseAugur Editorial · Summary by gemini-2.5-flash-lite from 1 source

MiniMax has released its M2.5 model, which is now open-source and trained using reinforcement learning for tasks like coding and tool use. The company highlights its cost-effectiveness, claiming it can run for $1 per hour at 100 tokens per second, making self-hosting feasible. The release also includes details about their 'Forge' RL training system, and early user feedback suggests it's viable for multi-turn interactions despite being token-hungry. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

RANK_REASON Open-source model release from a non-frontier lab with benchmark results.

Read on Smol AINews →

MiniMax M2.5 open-sources agent-native RL model, rivaling Sonnet on coding

COVERAGE [1]

  1. Smol AINews TIER_1 ·

    MiniMax-M2.5: SOTA coding, search, toolcalls, $1/hour

    **MiniMax-M2.5** is now open source, featuring an "agent-native" reinforcement learning framework called **Forge** trained across **200k+ RL environments** for coding, tool use, and workflows. It boasts strong benchmark scores like **80.2% SWE-Bench Verified** and emphasizes cost…

RELATED TOPICS