DeepSeek V4 Flash is the fast, economical V4 model: a 284B-total, 13B-active MoE with a 1M-token context window, thinking and non-thinking modes, tool use, and current regular pricing from $0.0028 per 1M cache-hit input tokens.
V4 Flash has 284B total parameters while activating 13B per token. DeepSeek positions it as the speed-and-economy member of V4, with reasoning close to Pro and comparable results on simple agent tasks.
Use Chat Completions for standard SDKs, the Anthropic-compatible endpoint for Claude Code, or the Responses endpoint for current Codex CLI. MyDeepSeek keeps the model identifier consistent across those routes.
Current regular rates separate cache hits from misses: $0.0028 versus $0.14 per 1M input tokens, with output at $0.28. Reused stable prefixes can therefore matter more than trimming a few output tokens.
It is DeepSeek's fast, economical V4 model, using a 284B-total and 13B-active mixture-of-experts architecture with a 1M-token context window.
Yes. The current model supports thinking and non-thinking modes, JSON output, tool calls, and agent workflows.
Use Chat Completions for general OpenAI-compatible clients, Anthropic Messages for Claude Code, and Responses for current Codex CLI integrations.
