DeepSeek V4.1 Flash specs, prices, where it runs and its benchmark standing

Name: DeepSeek V4.1 Flash
Author: DeepSeek

open weights · #35 overall, podium score 2.1 · #1 on 1 board · top 10 on 3 of 11 · released Sep 9, 2026

Provider: DeepSeek
Release date: Sep 9, 2026; Boards
Parameters: 763B total; Hugging Face
Modalities: Text, Image to Text; OpenRouter
License: MIT; Hugging Face
Context window: 1.05M tokens; Pricing
Pricing: $0.04 in / $0.3 out per 1M tokens; All prices
Hosting: 39 priced API routes · OpenRouter · self-host: weights on Hugging Face; Pricing
Hardware: Estimate: about 765 GB of weights as published (I8), about 763 GB at 8-bit, about 382 GB at 4-bit, plus memory for the context; Hugging Face config
Weights: Open weights
Paper: DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression; arXiv
Benchmarks: 11 boards captured, #35 overall

Rank	Board	Category	Score	Of
1	OpenRouterUsage · as “deepseek/deepseek-v4.1-flash-20261017”	Usage	56.0T tok	18
3	AA PricePrice · as “DeepSeek V4.1 Flash 1017 (Max)”	Price	$0.53	18
8	LiveBenchReasoning · as “deepseek-v4.1-flash-max”	Reasoning	81.1%	64
14	WebDev ArenaCoding · as “DeepSeek-V4.1-Flash-20261017 (Max)”	Coding	1,665	82
16	AA IntelligenceReasoning · as “DeepSeek V4.1 Flash 1017 (Max)”	Reasoning	32.6	50
19	Epoch ECIReasoning · as “DeepSeek V4.1 Flash 1017”	Reasoning	159.3	100
19	ARC-AGI-2Reasoning · as “DeepSeek V4.1 Flash 1017 (Max)”	Reasoning	78.1%	92
21	GPQA DiamondReasoning · as “deepseek-v4.1-flash-1017_max”	Reasoning	93.2%	100
25	AA SpeedSpeed · as “DeepSeek V4.1 Flash 1017 (max)” Unverified	Speed	14 tok/s	43
32	Text ArenaChat · as “DeepSeek-V4.1-Flash-20260910 (Max)”	Chat	1,475	84
34	Vision ArenaVision · as “DeepSeek-V4.1-Flash-20260910 (Max)”	Vision	1,263	85