DeepSeek V4.1 Flash 1017

The October 17, 2026 snapshot of DeepSeek V4.1 Flash supports text and image input, tool calling, and optional reasoning with low, high, and max effort. It has a shared 1,048,576-token context window and supports up to 393,216 output tokens within the remaining context. Released under the MIT license.

Reasoning
Vision
Tool Calling

Added Oct 17, 2026

Model weights

Pricing

Auto routing · per 1M tokens

Input: $0.100
Output: $0.60
Cache read: $0.0030

Compare provider prices

Specifications

Context window: 1M
Max output: 393.2K
Avg output (7d): 359 tokens; Longer than 37% of models

Benchmarks

Artificial Analysis

Design Arena

Sourced from Artificial Analysis.

Intelligence Index

24.7

Better than 77% of models compared

Agentic work

AutomationBench-AA

Workflow automation with guardrail penalties

47.8%

Better than 63% of models compared

AutomationBench-AA Tasks Completed

Fully completed workflows without guardrail violations

14.2%

Better than 58% of models compared

AA-Briefcase v1.1

Agentic knowledge work (Elo)

1118 Elo

Better than 57% of models compared

GDPval-AA v2.1

Economically valuable tasks (Elo)

1341 Elo

Better than 69% of models compared

Document reasoning

GDP.pdf

Professional PDF reasoning: all-pass rate

3.8%

Better than 22% of models compared

AA-LCR v1.1

Long context reasoning with updated grading

55.3%

Better than 47% of models compared

Reasoning

HLE

Humanity's Last Exam

10.8%

Better than 51% of models compared

Coding

Terminal-Bench v4.0

Practical coding and terminal tasks

5.6%

Better than 55% of models compared

SciCode

Python programming for scientific computing

35.5%

Better than 13% of models compared

Legacy benchmarks

AA-LCR (unversioned / legacy)

Long context reasoning evaluation