Skip to content

Benchmarks

Sources
8
Entries tracked
22,318
Last sync
17 min
All sources syncing normally
  • Arena · Text17 minutes ago
  • Arena · WebDev17 minutes ago
  • Arena · Vision17 minutes ago
  • Arena · Agent17 minutes ago
  • GPQA Diamond54 minutes ago
  • SWE-bench54 minutes ago
  • Terminal-Bench54 minutes ago
  • ARC-AGI-254 minutes ago

deepseek/deepseek-v3.2

45±0

44interval45

DeepSeekswing #76–#583

Overall ×1WebDev Arena ×1Vision Arena ×1Factuality Arena ×1Search Arena ×1Document Arena ×1Agent Arena ×1GPQA Diamond ×1AIME (math) ×1FrontierMath ×1SimpleQA Verified ×1ARC-AGI-2 ×1Humanity's Last Exam ×1SimpleBench ×1LiveBench ×1BFCL v4 (function calling) ×1dormantSWE-bench Verified (live board) ×1Terminal-Bench 4.0 ×1MMLU-Pro (Artificial Analysis) ×1LiveCodeBench (Artificial Analysis) ×1IFBench (Artificial Analysis) ×1HHEM Hallucination (Vectara) ×11 of 1034 models
ModelOrgIntervalSwingiiiiiiiiiiiiiiiiiiiiiiAdd to compare
583deepseek/deepseek-v3.2DeepSeek——44.8 29.2Ranked 44.8 = (1 × 29.2 + 3 × 50) ÷ 4. 29.2 is the mean percentile over the 1 of 22 selected axes this model is measured on; fewer axes means a stronger pull toward 50.——1/22low#76–583———————————29——————————

Blend is coverage-shrunk; the raw composite is shown greyed beside it · A low badge means fewer than 3 boards scored this model

The few things in AI and quantum that actually mattered this week — honestly summarised.

One email a week. Unsubscribe in one click. No tracking, no sharing.