Future AI Pulse
The latest open model rankings for October 2025 reveal a highly competitive race among the world’s top-performing large language models. New contenders have entered the leaderboard, showing how fast open-source AI is evolving across research labs and companies.
At the top of the list is GLM-4.6, developed by Zai and MIT, leading with an Arena Score of 1422. It’s followed closely by DeepSeek-V3.2-Exp-Thinking, another MIT-backed model, scoring 1419, and Qwen-3-235b-A22B-Instruct from Alibaba, with 1418.
The competition among these models highlights the growing sophistication of open architectures, many of which now rival or surpass closed-source alternatives in reasoning, language generation, and instruction following.
Rounding out the top ten are several notable names:
Kimi-K2-0905-Preview from Moonshot Longcat-flash-chat from Meituan MiniMax-M1 from MiniMax Gemma-3-27b-it from Google Mistral-Small-2506 from Mistral Command-A-03-2025 from Cohere gpt-oss-120b from OpenAI
The close margin of performance among these models — with scores separated by less than 80 points — underscores how quickly the open-source ecosystem is advancing. Many of these models are built on permissive licenses like Apache 2.0, enabling developers to fine-tune and deploy them freely.
As open AI development accelerates, research groups and startups are proving that community-driven innovation can compete head-to-head with corporate AI giants. This month’s leaderboard is a reminder that the future of AI is increasingly being built in the open.