Open R1 Update #4: DeepSeek-V3-0324 released with MIT license and improved code/math capabilities

AI Impact Summary

Open R1 Update #4 introduces DeepSeek-V3-0324, an updated base model for the R1 reasoning stack, now licensed under MIT. The release emphasizes instruction-following, coding, and math, and benchmarking suggests parity with GPT-4.5 and superiority over Claude-Sonnet-3.7 on several tasks, with targeted gains in front-end development, code execution, and multi-turn rewriting. Access pathways include Hugging Face Inference Providers, Text Generation Inference, and SG Lang, signaling an operational shift for teams to test and benchmark in their deployment pipelines.