⚡AI Agents Weekly: o3-mini, Qwen2.5-1M, Mistral Small 3, Janus-Pro, open-r1

o3-mini, Qwen2.5-1M, Mistral Small 3, Janus-Pro, open-r1

In today’s issue:

• OpenAI releases o3-mini

• Qwen launches Qwen2.5-1M & Qwen2.5 VL

• Improving RAG through multi-agent RL

• Mistral AI launches Mistral Small 3

• AI Agent Evaluation Report

• DeepSeek releases Janus-Pro

• HuggingFace works on a fully open reproduction of DeepSeek-R1

• Quantized version of DeepSeek-R1, top AI dev news, and much more.

---

---

Top Stories

OpenAI releases o3-mini

OpenAI launches o3-mini, their newest and most cost-efficient reasoning model, now available in both ChatGPT and API. The model excels in STEM capabilities, particularly in science, math, and coding, while maintaining the low cost and reduced latency of o1-mini. o3-mini introduces key developer features including function calling, Structured Outputs, and developer messages, making it production-ready from launch. Users can choose between three reasoning effort options—low, medium, and high—to optimize for specific use cases.

Experiments show that o3-mini delivers impressive performance improvements over its predecessor. With medium reasoning effort, it matches o1's performance in math, coding, and science while providing faster responses. Expert testers preferred o3-mini's responses to o1-mini 56% of the time and observed a 39% reduction in major errors. The model also demonstrates significant speed improvements, delivering responses 24% faster than o1-mini with an average response time of 7.7 seconds compared to 10.16 seconds.

Blog | System Card