o3-mini launches, OpenAI on "wrong side of history"

**OpenAI** released **o3-mini**, a new reasoning model available for free and paid users with a "high" reasoning effort option that outperforms the earlier **o1** model on STEM tasks and safety benchmarks, costing **93% less** per token. **Sam Altman** acknowledged a shift in open source strategy and credited **DeepSeek R1** for influencing assumptions. **MistralAI** launched **Mistral Small 3 (24B)**, an open-weight model with competitive performance and low API costs. **DeepSeek R1** is supported by **Text-generation-inference v3.1.0** and available via **ai-gradio** and replicate. The news highlights advancements in reasoning, cost-efficiency, and safety in AI models.