DeepSeek R1-0528

DeepSeek R1-0528 is DeepSeek's last dedicated open reasoning model (671B/37B MoE) with top-tier AIME math and a family of practical small distills.

Best for

• The best standalone open reasoner if you want a dedicated reasoning model
• Strong AIME math (~91) and GPQA Diamond (~81)
• Six distilled variants (1.5B–70B) for cheap local reasoning

How it compares — and the India angle

• The 8B/70B distills are the practical pick for cheap, local reasoning on Indian hardware.
• MIT licence allows commercial use and distillation — good for data-residency-constrained Indian firms.
• Reasoning is now subsumed into V4's thinking mode; R1 remains the choice only if you want a dedicated reasoner.