NEW · API PLANS - 23 models, one key, 125M - 2B tokens a month · from $30/mo See plans →
Home › Models › DeepSeek › DeepSeek V4.1 Flash 1017

DeepSeek V4.1 Flash 1017

High-throughput 1017 coding workhorse (90.6% Terminal-Bench 2.1, 384K max output).

High-speed October 1017 coding checkpoint combining 90.6% Terminal-Bench 2.1 accuracy, 384K maximum output length, and 1M context for multi-file SWE agents.

Built specifically for agentic coding loops, large patch synthesis, and shell workflows. Handles whole-file edits cleanly thanks to its 384K output ceiling while sustaining rapid streaming throughput.

Multi-file codingTerminal agentsLarge diff synthesisFast reasoningAgentic CLI codingLarge file generationTest fixingCI repair

Strengths

Verified 90.6% Terminal-Bench 2.1, 90.9% GPQA, 3471 Codeforces rating, huge 384K output limit, fast streaming speed.

Considerations

Default high-speed SWE sub-agent lane for OpenClaw, DeepSeek Harness, Cline, and Roo Code.

Input$0.3/1M
Output$1.2/1M
Cached read$0.007/1M
ReleasedOct 17, 2026

Live throughput history

source: status.ebotservers.com
Parameters671B
Context1M
Max output384K
Intelligence index32
TPS avg 24h59.9
TTFB p5024453 ms
Uptime100.0%
Published benchmarks
Terminal-Bench 2.1 90.6%
GPQA Diamond 90.9%
Codeforces 3471rating

Call it


 
 

Other models from DeepSeek