in the loop

DeepSeek V4 Flash 0731
#35 of 274 scored.

DeepSeekJul 31, 2026Open weights

Capabilities Index

Range 152.1–156.3

154.5

Context

Takes text

1.05M

Input

Per 1M tokens

$0.006

Output

Per 1M tokens

$1.28

Benchmarks

How it scores.

  • GPQA Diamond88.0%

    28th of 186

  • FrontierMath57.5%

    37th of 81

  • ARC-AGI-261.4%

    22nd of 79

  • SimpleQA Verified33.6%

    55th of 77

  • Mock AIME94.4%

    37th of 176

Everything else Epoch has for it

  • ARC-AGI89.0%
  • DTBench84.9%
  • WeirdML63.0%
  • ProofBench56.0%
  • SimpleBench53.3%
  • LMCA49.1%
  • Chess Puzzles29.5%
  • Mystery Game Puzzles27.3%
  • FrontierMath-Tier-4-v2-Private24.4%
  • FrontierCode18.8%

On the index

Around it.

  1. #32DeepSeek V4.1 FlashDeepSeek · $0.30 in154.9
  2. #33Muse Spark 1.2Meta · $1.25 in154.9
  3. #34Gemini 3.1 ProGoogle154.8
  4. #35DeepSeek V4 Flash 0731DeepSeek · $0.006 in154.5
  5. #36Gemini 3.5 FlashGoogle · $1.50 in154.5
  6. #37Gemini 3.6 FlashGoogle · $0.75 in154.3
  7. #38Muse Spark 1.1Meta · $1.25 in154.2

Sources