← All models

DeepSeek-V4.1-Flash

Published Publisher DeepSeekdeepseek-flash

Overview

Released on September 10, 2026, DeepSeek-V4.1-Flash is an efficiency-oriented, multimodal model with native visual understanding. It uses a 552-billion-parameter mixture-of-experts architecture with an asymmetric causal encoder–decoder: approximately 8 billion parameters are active for input and 16 billion for output. DeepSeek says its KV cache requires one-quarter of the previous generation’s HBM and one-eighth of its SSD storage. (deepseek.com)

Capabilities and uses

The model supports text and image understanding and is intended for reasoning, coding, and agentic workflows. DeepSeek provides open model weights under the MIT license. Its model card describes training with a context extended to 1 million tokens. Benchmark results depend on the evaluation settings and agent framework used. (huggingface.co)

API version and limitations

The API identifier deepseek-flash is an alias, not a fixed snapshot. As of November 16, 2026, DeepSeek’s pricing documentation identifies the served version as DeepSeek-V4.1-Flash-1017, an update announced on October 17, 2026. The current API documentation lists a 1-million-token context, a maximum output of 384,000 tokens, vision, tool calls, JSON output, and thinking and non-thinking modes. Fill-in-the-middle completion is limited to non-thinking mode. These current API limits and features should not automatically be attributed to the original September snapshot. (api-docs.deepseek.com)

Self-hosting requires substantial resources: in the original announcement, DeepSeek invited operators with 2,000 GPUs and a storage cluster to discuss large-scale deployment. Outputs should be checked for accuracy before consequential use. (deepseek.com)

Official API pricing

Queried November 16, 2026. The following current prices apply to the deepseek-flash API alias, which the official pricing pages identify as DeepSeek-V4.1-Flash-1017. They must not be assumed to be the historical price of the original September snapshot; its exact launch prices could not be verified from the accessible announcement text. (api-docs.deepseek.com)

All amounts below are per 1 million tokens. Input is billed differently for cache hits and misses; output is billed separately.

Official price schedule Period Input, cache hit Input, cache miss Output
International, USD Peak $0.006 $0.30 $1.20
International, USD Off-peak $0.003 $0.15 $0.60
Mainland China, CNY Peak ¥0.04 ¥2 ¥8
Mainland China, CNY Off-peak ¥0.02 ¥1 ¥4

Peak hours are Monday–Friday, 01:00–04:00 and 06:00–10:00 UTC, excluding Chinese public holidays. All other hours are off-peak, including weekends and Chinese public holidays. The CNY and USD schedules are separately published prices, not currency conversions. Additional tool usage can incur separate charges. (api-docs.deepseek.com)

Official sources