Best LLMs for Science — DeepSeek
The best DeepSeek model for science is DeepSeek V4 Flash 0731, scoring 90.8% on GPQA Diamond — graduate-level physics, chemistry and biology. DeepSeek V4 Flash 0423 (90.8%) and DeepSeek V4 Pro (88.8%) round out the top three.
- 1
deepseek-v4-flash-0731ReasoningToolsJSON49.9 intel · $0.090/M · 101 t/s90.8%GPQA - 2deepseek-v4-flashReasoningToolsJSON49.9 intel · $0.140/M · 114 t/s90.8%GPQA
- 3deepseek-v4-proReasoningToolsJSON44.3 intel · $0.435/M · 67 t/s88.8%GPQA
- 4deepseek-r1ReasoningToolsJSON20.1 intel · $0.700/M · 164K ctx81.3%GPQA
- 5deepseek-v3.2ReasoningToolsJSON24.7 intel · $0.269/M · 50 t/s75.1%GPQA
- 6deepseek-v3.1-terminusReasoningToolsJSON21.4 intel · $0.270/M · 53 t/s75.1%GPQA
- 7deepseek-r1-distill-llama-70bReasoning9.9 intel · $0.800/M · 8K ctx40.2%GPQA
How this is ranked
AI models ranked by GPQA Diamond — graduate-level physics, chemistry and biology questions written to be un-Googleable, so the score reflects reasoning from knowledge rather than retrieval. The best large language models for scientific research and hard-science work.
Frequently asked
What is the best DeepSeek model for science?
The best DeepSeek model for science is DeepSeek V4 Flash 0731, scoring 90.8% on GPQA Diamond — graduate-level physics, chemistry and biology. DeepSeek V4 Flash 0423 (90.8%) and DeepSeek V4 Pro (88.8%) round out the top three.
What's a good alternative to DeepSeek V4 Flash 0731?
DeepSeek V4 Flash 0423 (90.8%) is the closest alternative on this metric, followed by DeepSeek V4 Pro (88.8%). See the full ranking above for the tradeoffs.
How many DeepSeek models are there?
modelgrep tracks 12 DeepSeek models with live benchmarks, speed, latency and per-provider pricing, led on intelligence by DeepSeek V4 Flash 0731. 7 of them qualify for this ranking.