deepseek-v4-flash vs deepseek-v4-pro speed comparison

Based on 89 anonymous user runs.

Verdict: deepseek-v4-flash has faster output (median 142 vs 81 tok/s); deepseek-v4-flash has faster TTFT (0.71s vs 1.41s).
Share and embed

Post to social channels, or use Markdown and badges for GitHub/README.

XFacebook微博LinkedIn
[![deepseek-v4-flash is faster than deepseek-v4-pro: 142 tok/s on TOKRACE](https://tokrace.com/api/badge/compare/deepseek-v4-flash-vs-deepseek-v4-pro?locale=en)](https://tokrace.com/en/compare/deepseek-v4-flash-vs-deepseek-v4-pro)
Median output tok/s14281
Average output tok/s13887
TTFT0.71s1.41s
Peak tok/s346293
Samples5633

· Data comes from voluntary anonymous sharing; medians reduce jitter · Updates every 5 minutes

· Speed is affected by network, time of day and provider load · Methodology

How to use this comparison

Writing/long output: Prioritize median output tok/s and peak speed.

Chat/agents: TTFT usually has a bigger UX impact.

Model selection: Rerun your real Prompt and inspect output quality too.

Run with current dataView full leaderboard

FAQ

Which model outputs faster, deepseek-v4-flash or deepseek-v4-pro?

deepseek-v4-flash has faster output (median 142 vs 81 tok/s); deepseek-v4-flash has faster TTFT (0.71s vs 1.41s).

Why can output speed and TTFT have different winners?

Output tok/s measures sustained generation speed, while TTFT measures the wait until the first token. A model can generate long text faster while still taking longer to start.

How should I rerun this comparison?

Use the arena with the same Prompt, temperature and network conditions, then repeat a few times and combine the speed data with output quality.

Can I embed this comparison in GitHub or an article?

Yes. This page provides Markdown and HTML badges. The badge image URL is https://tokrace.com/api/badge/compare/deepseek-v4-flash-vs-deepseek-v4-pro?locale=en.