System Card← leaderboard
DeepSeek-AI

DeepSeek Flash

#12of 28 · general

DeepSeek V4.1 Flash 1017

763Bparams · MoE
1Mcontext
45Ttraining tokens
2026-10-17released
weights ↗repo ↗
50100Coding 58 · #9 of 28Agents 51 · #11 of 29Cyber: no readingEngineering 39 · #14 of 26Math 31 · #17 of 26 · one sourceHealth: no readingLegal 50 · #11 of 27Professional 35 · #16 of 28Writing: no readingInstruct: no readingHallucination 15 · #23 of 26 · one sourceMultimodal 49 · #11 of 24 · one source
Coding58#9
Agents51#11
Cyber—
Engineering39#14
Math31#17
Health—
Legal50#11
Professional35#16
Writing—
Instruct—
Hallucination15#23
Multimodal49#11

History

20%40%60%80%100%Apr 2025Jul 2025Oct 2025Jan 2026Apr 2026Jul 2026Oct 2026DeepSeek V3 · no price on record · 2025-03-24 · still served · OTIS Mock AIME 2024-2025 26.8% fractional score (0–1 scale)DeepSeek R1 · $0.7/M in · 2025-05-28 · still served · OTIS Mock AIME 2024-2025 59.9% fractional score (0–1 scale)DeepSeek V4 Flash 0731 · $0.0144/M in · 2026-07-31 · still served · OTIS Mock AIME 2024-2025 94.4% fractional score (0–1 scale)DeepSeek V4.1 Flash · no price on record · 2026-09-09 · still served · OTIS Mock AIME 2024-2025 98.3% fractional score (0–1 scale)DeepSeek V4.1 Flash 1017 · $0.07/M in · 2026-10-17 · still served · OTIS Mock AIME 2024-2025 99.4% fractional score (0–1 scale)V4.1 Flash0731R1V3
  • DeepSeek-AI
  • older → newer
$ / M input tokens5 of 176 priced models · price is a third-party claim, not a vendor quote

Usage

163TtokensSep 10 – Oct 6▼ 11% 7d
50T100T150T200TSep 10Oct 1

Geography

China25.0%
United States12.6%
Japan5.8%
Germany3.8%
Singapore3.7%
Brazil3.6%
Hong Kong3.4%
India2.6%
France2.4%
Vietnam2.3%

Measured

≈ imputed · amber dot = lab-self-reported among the sources
GPQA Diamond93.2%#8Epoch AI
HLE42.6%#19LLM Stats
Vals Index61.3#7Vals.ai

Neighbors

GLM→ more factual · stronger professionalQwen Max→ more factual · stronger professionalMiMo Pro→ weaker multimodal · more factualGemini Flash→ weaker coding · more factualClaude Haiku→ weaker agent