in the loop

DeepSeek-V3.2
#76 of 274 scored.

DeepSeekDec 1, 2025Open weights

Capabilities Index

Range 144.5–147.4

146.3

Context

Takes text

164K

Input

Per 1M tokens

$0.26

Output

Per 1M tokens

$0.42

Benchmarks

How it scores.

  • GPQA Diamond77.9%

    70th of 186

  • ARC-AGI-24.0%

    57th of 79

  • Mock AIME87.8%

    59th of 176

  • Terminal-Bench39.6%

    20th of 35

Everything else Epoch has for it

  • DTBench76.0%
  • Aider polyglot74.2%
  • ARC-AGI57.0%
  • FrontierMath-2025-02-28-Private38.8%
  • LMCA33.9%
  • CL-bench12.4%
  • Chess Puzzles9.5%
  • ProofBench8.0%
  • CL-bench Life7.4%
  • FrontierMath-Tier-4-2025-07-01-Private3.5%

On the index

Around it.

  1. #73Qwen3.5 397B-A17BAlibaba · $0.55 in146.7
  2. #74Qwen3.6 27BAlibaba · $0.32 in146.5
  3. #75Grok 4xAI146.4
  4. #76DeepSeek-V3.2DeepSeek · $0.26 in146.3
  5. #77Nemotron 3 UltraNvidia · $0.50 in146.2
  6. #78DeepSeek-V4-FlashDeepSeek · $0.006 in146.1
  7. #79Kimi K2 ThinkingMoonshot · $0.60 in146.0

Sources