Skip to content

DeepSeek

DeepSeek V4.1 Flash (Oct 2026)

Second-highest open-weights model on the Epoch Capabilities Index (159.3, listed there as “DeepSeek V4.1 Flash 1017”), behind Kimi K3.1 Pro (162.0), as read on 10 November 2026.

Announced
Access
Public

Benchmark scores

TestScoreBest on the trackerHow it was measured
Epoch Capabilities Indexmany tests combined into one score159 pointsGPT-6.2 Astra 174 pointsIndependentlisted as "DeepSeek V4.1 Flash 1017", CI 157.08 to 161.86Run 10 Nov 2026
APEX-Agentsprofessional office tasks70%Gemini 4 Carbon 84%Independentpass@1, max effort; listed as "DeepSeek-V4.1-Flash-1017"
ARC-AGI-3puzzle games, standard harnessTypical first-time human player: 100%0.2%Claude Fable 5.2 81%IndependentARC Prize config "DeepSeek V4.1 Flash 1017 (Max)" (deepseek-v4-1-flash-1017-max), standard harness
FrontierMath T1–3hard maths problems71%Claude Fable 5.2 95%Independentmax reasoning effort (deepseek-v4.1-flash-1017_max)Run 19 Oct 2026
GPQA DiamondPhD-level science questionsPhD experts: 70%Saturated93%Gemini 4 Argon 96%Independentmax reasoning effort (deepseek-v4.1-flash-1017_max)Run 19 Oct 2026
FrontierMath T4research-level maths problemsSaturated46%GPT-6.1 Sol 100%Independentmax reasoning effort (deepseek-v4.1-flash-1017_max)Run 19 Oct 2026
ARC-AGI-2visual puzzles, model aloneAverage person: 60%Saturated78%GPT-6 Astra 95%IndependentARC Prize config "DeepSeek V4.1 Flash 1017 (Max)" (deepseek-v4-1-flash-1017-max)
OTIS Mock AIMEcompetition mathsSaturated99%GPT-5.5 100%Independentmax reasoning effort (deepseek-v4.1-flash-1017_max)Run 19 Oct 2026

Compare every model on the same tests →

Sources