DeepSeek-V4.1-Flash-1017

Available

DeepSeekOpen source

DeepSeek-V4.1-Flash-1017 is a refreshed checkpoint of DeepSeek-V4.1-Flash released 2026-10-17, with improved reasoning, coding, agent tasks and visual understanding, and shorter task execution / output length to cut cost. Same architecture as V4.1-Flash: 552B-total multimodal MoE (about 8B active per input token / 16B per output token), 1,048,576-token context, text + native image understanding. MIT-licensed open weights on Hugging Face and ModelScope, with inference code and deployment instructions. API pricing $0.30 / $1.20 per 1M input/output at peak, $0.15 / $0.60 off-peak.

Official page ↗Model card ↗

Specifications

License: Open source · MIT
Weights: Downloadable
Architecture: Mixture-of-Experts
Parameters: 552B · 16B active
Context window: 1.0M tokens
Max output: —
Knowledge cutoff: —
Price (in / out, $/M): $0.3 / $1.2
Modalities: TextVisionCode

Benchmarks

No benchmark scores recorded yet. Spotted some? Submit a correction.

Vendor-reported figures are claims until independently verified. See methodology.