DeepSeek V4.1 Flash

NEW ARCHITECTURE FLAGSHIP

Released Sept 10, 2026; first of DeepSeek's Causal Encoder-Decoder family with native image+text, 1M context and 384K output. Replaces V4-Flash and V4-Pro from Sept 14.

Context

1,000,000 tokens

Architecture

552B MoE (8B / 16B Activated)

Official API (1M)

$0.3 in / $1.2 out (Open)+ Weights Free to Self-Host

License

MIT License

Published benchmarks · lab-reported

Vendor-published figures, not independently reproduced. Sources in Verification below.

SWE-BENCH

74.2% (DeepSWE v1.1)

TERMINAL-BENCH

90.6%

MMLU-PRO

74.1%

GPQA DIAMOND

90.9%

Quickstart

curl -s https://modelregistry.tirup.in/api/cli?model=deepseek-v4-1-flash

OFFICIAL ANNOUNCEMENT RESEARCH PAPER WEIGHTS / REPOSITORY API DOCS

BACK TO LEDGER

Sources

DeepSeek V4.1 Flash announcement · DeepSeek
DeepSeek V4.1 Flash API reference · DeepSeek
DeepSeek V4.1 Flash open weights · DeepSeek
DeepSeek V4.1 Flash technical report · DeepSeek

Report an error in this record