Return to List
DeepSeek V4.1 Flash
Model ID:
online
Pricing
Input Token Cost 0.085 credits
Output Token Cost 0.3 credits
Cached Token Cost 0.02 credits
Model Details
Context Tokens 1048576
Maximum Output 1048576
Prompt Format DeepSeek
Architecture DeepSeekV4.1
Parameters 763B
Source Link
Tags Assistant, Tools, Coding, Smart
Recent Metrics (Past 4 hours)
Avg Input Speed[1] 5750.8 tokens/sec
Avg Output Speed[1] 147.4 tokens/sec
Description

DeepSeek V4.1, a significant leap forward in inferencing speed while sharing v4 Flash's excellent response quality.

Prompt Formatting

If you use /v1/completions, formatting your prompts according to the model's instruct format can greatly improve results. If you use /v1/chat/completions, this is not relevant.

A properly formatted prompt would appear as follows, with non-formatting example content shown in (italicized parenthesis):

Showing "DeepSeek" formatting.
DeepSeekV4's chat format is not compatible with a simple Jinja template

[1] Aggregate performance metrics should only be used as an estimate.
Speeds can be affected by model traffic, sampler choice, prompt prefix re-use, and many other things.

Data fetched 2026-11-16 19:12:23 UTC. Click to refresh.

Mancer AI 2025 - Discord - Contact Us (JavaScript required!)
Terms of Use - Privacy Policy