DeepSeek V4.1 Flash

Name: DeepSeek V4.1 Flash
Author: DeepSeek

Released

2026-09-10

Last refreshed

2026-10-08

Status

Researched 32d ago

Open sourceCommercial use: permittedMultimodalRAGAgentsLong contextVisionJSON / Tool use

DeepSeek releases · 7 in the last 12 months · this family litChangelog →

Specifications

Family: DeepSeek V4.1
Released: 2026-09-10
Context: 1M
Max output: 384K
Parameters: 552B
Architecture: Mixture of Experts
Specialization: general
Openness: Open source
License: MITOSI-approvedCommercial use: permitted
Weights: Available
Code: Unknown
Training: Pretrained

Created by

DeepSeek

Advancing artificial general intelligence (AGI).

Hangzhou, Zhejiang, China

Founded 2023

Website

Pricing

Output / 1M

$0.600

Input / 1M

$0.150

Cheapest of 2 routes · DeepSeek Platform · cache read $0.003

Providers(2)

DeepSeek Platform OpenRouter

View 2 provider routes

Links

Website HuggingFace

About

DeepSeek V4.1 Flash is a 552B-parameter (8B prefill / 16B decode activated) mixture-of-experts model with Compressed Expert Dispatch (CED), released September 10, 2026 under the MIT license. It supports 1M-token context, up to 384K output tokens, multimodal vision input, reasoning, function calling, tool use, structured outputs, and prompt caching. Weights are available on Hugging Face. The DeepSeek API serves it as deepseek-flash; legacy API aliases deepseek-v4-flash and deepseek-v4-flash-vision-exp route to this model. Off-peak API pricing: $0.15/1M input, $0.60/1M output (cache read: $0.003/1M); peak hours are 2×.

Providers(2)

Compare all 2

Compare API pricing across 2 tracked provider routes for input and output tokens, batch, and cached reads when available.

Provider	Input / 1M	Output / 1M	Cache	Route
DeepSeek Platform	$0.150	$0.600	read $0.003	Serverless
OpenRouter	$0.150	$0.600	read $0.003	Serverless

Available via routers & gateways(4)

LiteLLM

Gateway

Open-source Python SDK and proxy server that unifies 100+ LLM APIs behind a single OpenAI-compatible interface, with load balancing, cost tracking, and configurable failover.

Free OSSDeepSeek Platform

OpenRouter

Hybrid

Unified hybrid gateway to 400+ models from 60+ providers via a single OpenAI-compatible API, with optional auto-routing that selects the best model per prompt.

PassthroughDeepSeek Platform

Requesty

Hybrid

AI gateway to 400+ LLM providers with intelligent routing, caching, guardrails, and governance; flat 5% markup on model costs with no subscription fee.

Passthrough + feeDeepSeek Platform

Respan

Hybrid

Unified LLM engineering platform (gateway + observability + evals + prompt management) routing across 250+ models; previously Keywords AI, rebranded February 2026.

SubscriptionDeepSeek Platform

Capabilities

VisionMultimodalReasoningJSON / Tool useStructured OutputsPrompt Caching

Benchmark peer barsfor RAG

No task-mapped benchmark peers are available for this model yet.

Benchmark scores(19)

Scores are benchmark-specific and are direction-aware: the same numeric gap can mean very different outcomes across suites. Use the leaderboard context and this model's provider route to decide whether the winning margin is meaningful for your workload.

Benchmark	Score	Version	Evaluation	Source
Google-Proof Q&A	90.9	GPQA Diamond Pass@1; max reasoning effort (reasoning_effort=100)Observed 2026-09-10	—	Source
Humanity's Last Exam	36.8	HLE Pass@1; text-only; max reasoning effortObserved 2026-09-10	—	Source
Humanity's Last Exam — With Tools	63.9	HLE with tools Pass@1; max reasoning effortObserved 2026-09-10	—	Source
Codeforces	3471.0	Codeforces Rating; max reasoning effortObserved 2026-09-10	—	Source
MathArena Apex	65.6	MathArena Apex Pass@1; max reasoning effortObserved 2026-09-10	—	Source
Terminal-Bench 2.1	90.6	Terminal-Bench 2.1 Pass@1; DeepSeek Harness Minimal mode; 1M contextObserved 2026-09-10	—	Source
Terminal-Bench 3.0	30.0	Terminal-Bench 3.0 Pass@1; DeepSeek Harness Minimal mode; 1M contextObserved 2026-09-10	—	Source
Terminal-Bench 4.0	31.2	Terminal-Bench 4.0 Pass@1; DeepSeek Harness Minimal mode; 1M contextObserved 2026-09-10	—	Source
DeepSWE 1.1	74.2	DeepSWE 1.1 Resolved; mini-SWE harness; max reasoning effortObserved 2026-09-10	—	Source
ProgramBench Almost@1	20.3	ProgramBench Almost@1; DeepSeek Harness Minimal mode; 1M contextObserved 2026-09-10	—	Source
NL2Repo-Bench	65.4	NL2Repo-Bench Score; DeepSeek Harness Minimal mode; 1M contextObserved 2026-09-10	—	Source
CyberGym	88.1	CyberGym Pass@1; max reasoning effortObserved 2026-09-10	—	Source
SEC-Bench Pro	62.8	SEC-Bench Pro Pass@1; Claude Code harness; max reasoning effortObserved 2026-09-10	—	Source
ExploitGym	15.3	ExploitGym Pass@1; max reasoning effortObserved 2026-09-10	—	Source
AutomationBench	54.8	AutomationBench Pass@1; official scaffold; max reasoning effortObserved 2026-09-10	—	Source
Agents' Last Exam	31.8	Agent's Last Exam Pass@1; official scaffold; max reasoning effortObserved 2026-09-10	—	Source
Chartography w/ tools	78.9	Chartography w/ tools Pass@1; Claude Code harness; 512k contextObserved 2026-09-10	—	Source
BabyVision w/ tools	89.6	BabyVision w/ tools Pass@1; Claude Code harness; 512k contextObserved 2026-09-10	—	Source
ZeroBench-main w/ tools	49.0	ZeroBench-main w/ tools Pass@5; Claude Code harness; 512k contextObserved 2026-09-10	—	Source