Head to head

DeepSeek V4.1 Flash vs DeepSeek V4 Pro

DeepSeek V4.1 Flash scores higher on 3 of the 4 benchmarks both models publish. The widest gap is on Humanity's Last Exam, where DeepSeek V4 Pro leads by 5.9 points. Both have open weights; DeepSeek V4.1 Flash is far lighter to host, needing about 381 GB at 4-bit against 1.1 TB for DeepSeek V4 Pro.

DeepSeek V4.1 FlashDeepSeek · OpenLLM Buddy catalog79.8compositeDeepSeek V4 ProDeepSeek · Open weights91.0composite

Shared benchmarks

Raw published scores; the higher score on each row is highlighted. LMArena Elo is a human-preference vote and is not counted as a win.

  • GPQA Diamond
    90.990.1

    Pass@1, max reasoning effortV4-Pro (Preview) Max, pass@1

  • Humanity's Last Exam
    36.842.7

    Pass@1, no tools, max effort (text-only subset: 39.1)V4-Pro-0813, no tools

  • Humanity's Last Exam (with tools)
    63.960.0

    Pass@1, with tools, max effortV4-Pro-0813, with tools

  • Terminal-Bench
    90.687.9

    Terminal-Bench 2.1, Pass@1, max effortTerminal-Bench 2.1, V4-Pro-0813, max effort, DeepSeek Harness

  • LMArena Text Elo · not counted
    14761465

    LMArena text overall, 'DeepSeek-V4.1-Flash-20260910 (Max)', ±6, as of 2026-10-27DeepSeek-V4-Pro (High, 20260813), ±7, board dated 2026-10-27

Should you pick DeepSeek V4.1 Flash or DeepSeek V4 Pro?

Pick DeepSeek V4.1 Flash if…

  • You care most about Humanity's Last Exam (with tools), Terminal-Bench, where it scores at least 2 points higher.
  • People prefer its answers in blind chats: LMArena Elo 1476 against 1465.
  • You want the lighter model to host: about 381 GB at 4-bit against 1.1 TB.
  • You need fast responses: about 217 tokens per second through APIs, against 94.
  • Cost matters: $1.20 per million output tokens against $3.96.
  • You want it running in minutes on a dedicated GPU through OpenLLM Buddy, in an uncensored build.

Pick DeepSeek V4 Pro if…

  • You care most about Humanity's Last Exam, where it scores at least 2 points higher.

Hardware and cost to run

DeepSeek V4.1 Flash

Even at 4-bit it needs about 381 GB (4× H200 (564 GB)), well beyond a home workstation, so most people run it on rented multi-GPU servers. Full BF16 precision needs about 1.3 TB (Multiple nodes (over 1.1 TB)).

On OpenLLM Buddy: DeepSeek-V4.1-Flash-UNCENSORED-FP8 runs on 4× H200 141GB as an FP8 build, with an OpenAI-compatible API and no per-token fees. Prices start at $280 USD for 11 hours.

Deploy DeepSeek V4.1 Flash
Full GPU memory table for DeepSeek V4.1 Flash

DeepSeek V4 Pro

Even at 4-bit it needs about 1.1 TB (8× H200 (1,128 GB)), well beyond a home workstation, so most people run it on rented multi-GPU servers. Full BF16 precision needs about 3.7 TB (Multiple nodes (over 1.1 TB)).

Full GPU memory table for DeepSeek V4 Pro

Category profile

Category index uses every test each model reports, not only the shared ones.

ReasoningCodingAgenticKnowledge
DeepSeek V4.1 FlashDeepSeek V4 Pro
Reasoning75.986.7
Coding72.189.4
Agentic91.388.1
Knowledge—100.0

Specs side by side

DeepSeek V4.1 FlashDeepSeek V4 Pro
LabDeepSeekDeepSeek
WeightsOpenOpen
Released2026-09-102026-08-13
Parameters552B total, 16B active (MoE)1.6T total, 49B active (MoE)
Context1M1M
LicenseMITMIT
4-bit memory381 GB · 4× H200 (564 GB)1.1 TB · 8× H200 (1,128 GB)
API price$0.300 in / $1.20 out per 1M tokens$1.32 in / $3.96 out per 1M tokens
API speed217 tok/s94 tok/s

Related comparisons

Sources

ModelTestScoreSettingSource
DeepSeek V4.1 FlashGPQA Diamond90.9Pass@1, max reasoning effortDeepSeek-V4.1-Flash model card (Hugging Face)
DeepSeek V4 ProGPQA Diamond90.1V4-Pro (Preview) Max, pass@1Hugging Face model card (DeepSeek-V4-Pro)
DeepSeek V4.1 FlashHumanity's Last Exam36.8Pass@1, no tools, max effort (text-only subset: 39.1)DeepSeek-V4.1-Flash model card (Hugging Face)
DeepSeek V4 ProHumanity's Last Exam42.7V4-Pro-0813, no toolsHugging Face model card (DeepSeek-V4-Pro-0813)
DeepSeek V4.1 FlashHumanity's Last Exam (with tools)63.9Pass@1, with tools, max effortDeepSeek-V4.1-Flash model card (Hugging Face)
DeepSeek V4 ProHumanity's Last Exam (with tools)60.0V4-Pro-0813, with toolsHugging Face model card (DeepSeek-V4-Pro-0813)
DeepSeek V4.1 FlashTerminal-Bench90.6Terminal-Bench 2.1, Pass@1, max effortDeepSeek-V4.1-Flash model card (Hugging Face)
DeepSeek V4 ProTerminal-Bench87.9Terminal-Bench 2.1, V4-Pro-0813, max effort, DeepSeek HarnessHugging Face model card (DeepSeek-V4-Pro-0813)
DeepSeek V4.1 FlashLMArena Text Elo1476LMArena text overall, 'DeepSeek-V4.1-Flash-20260910 (Max)', ±6, as of 2026-10-27Arena (LMArena) text leaderboard
DeepSeek V4 ProLMArena Text Elo1465DeepSeek-V4-Pro (High, 20260813), ±7, board dated 2026-10-27LMArena (arena.ai) text leaderboard

Scores are published results from model cards, lab reports, and public leaderboards, each linked to its source. Settings such as thinking mode or test version differ between sources and are shown next to each score. OpenLLM Buddy catalog models are uncensored fine-tunes; their scores are those of the official base model. Data collected 2026-10-30. How the index works