大模型网摘deepseek-flash vs deepseek-v4-pro Quick Reference: Model Names, Peak/Off-Peak Pricing, 1M Context and Limits
The two current DeepSeek API model names have settled: deepseek-flash and deepseek-v4-pro. The former is served by the DeepSeek-V4.1-Flash-1017 snapshot, the latter by DeepSeek-V4-Pro-0813. The legacy names deepseek-v4-flash and deepseek-v4-flash-vision-exp are still accepted, but those models are retired — requests are handled by V4.1-Flash-1017 and billed at Flash prices. Apps still on the old names should switch soon.
Dual-format endpoints and key parameters
Both formats remain available: the OpenAI-compatible endpoint at https://api.deepseek.com and the Anthropic-compatible endpoint at https://api.deepseek.com/anthropic. Both models carry a 1M-token context length with 384K max output. In the capability matrix, JSON Output, Tool Calls, the Responses API and Chat Prefix Completion (Beta) are supported on both; FIM completion is non-thinking mode only; vision input is Flash-exclusive — V4-Pro does not accept images. As for thinking mode, Flash defaults to thinking and both modes can be switched.
Peak and off-peak pricing at a glance
Per the official pricing page (as of press time), per million tokens: deepseek-flash input cache hit is $0.003 off-peak and $0.006 peak; cache miss $0.15/$0.30; output $0.60/$1.20. The three tiers for deepseek-v4-pro are $0.022/$0.044, $0.66/$1.32 and $1.98/$3.96. Peak hours are 01:00-04:00 and 06:00-10:00 UTC, Monday through Friday, excluding Chinese public holidays; everything else is off-peak — including weekends and holidays — at half the peak rate. Our peak/off-peak explainer covers the mechanics.
Concurrency limits and Search & Fetch billing
Concurrency differs sharply: 2500 for Flash versus 500 for V4-Pro, so high-parallelism workloads need isolation and queueing planned ahead. The standalone Search and Fetch tools bill per successful request at $0.75 per thousand each; server-side calls inside Harness, Claude Code and Codex are currently free. Putting the numbers to work: on cache-miss input and output, Flash off-peak runs at roughly a quarter to a third of V4-Pro's price — if Flash covers your workload, there is little reason to pay for Pro; for Pro-grade coding and reasoning, schedule batches off-peak. Start with the Python quickstart if you are new, and see our tier comparison for how the two models differ.