Media & Culture

DeepSeek R2 just went open-source and it's matching GPT-4o on 9 of 12 benchmarks — for literally $0 in API costs

DeepSeek R2 just went open-source and it's matching GPT-4o on 9 of 12 benchmarks — for literally $0

Deep Dive

The benchmark sheet dropped this morning and people are losing it in the ML community. What DeepSeek R2 scores: •MMLU: 90.8 (GPT-4o: 88.7) •HumanEval coding: 93.2 — new open-source SOTA •MATH reasoning: 88.9 •Runs on a single A100, fully local, zero API costs Hugging Face hit 300k downloads in the f