Sign up
All models
DeepSeek V4 Flash

DeepSeek V4 Flash

DeepSeek

Official release of DeepSeek-V4-Flash, superseding the preview version, with substantially enhanced agentic capabilities. 304B parameters with a 1M token context window and a speculative decoding module attached for fast inference. Hosted via Fireworks.

Try on Simtheory
Qwen 3.6 Plus Frontier Claude 4.5 Sonnet Frontier Claude 4.5 Sonnet (thinking) Frontier Claude 4.6 Opus Frontier Claude 4.6 Opus Max Frontier Claude 4.6 Sonnet Frontier Claude 4.6 Sonnet Max Frontier Claude 4.7 Opus Frontier Claude 4.7 Opus Max Frontier Claude Fable 5 Frontier

Specifications

Context window 1,000,000 tokens
Max output 131,072 tokens
Knowledge cutoff March 2026
Multipart messages No
Vision capabilities No

Default Parameters

Temperature 0.3
Top P 1.0
Frequency penalty 0.0
Presence penalty 0.0
Top K 40

More from DeepSeek

DeepSeek V3.2

DeepSeek V3.2

Latest DeepSeek model hosted in the US.

163,840 ctx 32,000 out
DeepSeek-R1

DeepSeek-R1

Deep reasoning model from DeepSeek hosted securely in the USA

131,072 ctx 8,192 out
DeepSeek V4 Pro 0813

DeepSeek V4 Pro 0813

Frontier

DeepSeek's August 13 frontier Mixture-of-Experts model for long-context reasoning, coding, and agentic work, served through Fireworks AI with a 1M-token …

1,000,000 ctx 131,072 out