Machine translation

DeepSeek V4 Flash (via DeepSeek API)

Translation with DeepSeek's open-weight V4 Flash model on DeepSeek's first-party API in China, priced from token rates.

The estimate is $0.465 per million characters at peak rates, or roughly $0.23 off-peak. Actual billing is by tokens, so language and output length matter. DeepSeek's API processes data in China and has no documented DPA. The MIT-licensed model is also available from other hosts; check the region for this specific model rather than assuming every host offers EU processing.

Pricing

Published rates converted using workload assumptions. Read the notes. Token-to-character conversion assumes English source, short prompts and European-language output. Quality and retries are not priced. How to read these estimates.

Derived from peak-hour token prices: $0.30 per 1M input tokens (cache miss), $1.20 per 1M output tokens. 0.35 x $0.30 + 0.30 x $1.20 = $0.105 + $0.36 = $0.465 per 1M characters. Assumptions (shared by all LLM-derived rows): English source at ~4 characters per token, so 1M characters = 250k source tokens; a ~200-token prompt per 2,000-character chunk adds 500 x 200 = 100k tokens, giving 350k input tokens per 1M characters; output for European target languages ~1.2x the source = 300k tokens. Cost per 1M characters = 0.35 x input price + 0.30 x output price (per 1M tokens). Non-Latin target languages (CJK, Arabic, Hindi) and non-English sources can use noticeably more tokens. Off-peak hours, weekends and Chinese public holidays bill at 50% (~$0.23/M chars), and cached prompt prefixes cost $0.006/M tokens. Data is processed in China. The MIT-licensed weights are also served by third-party hosts (e.g. DeepInfra at $0.06/$0.18, Nebius in the EU at $0.14/$0.28 per 1M tokens).

Other machine translation comparisons

• DeepL API vs Google Cloud Translation
• DeepL API vs Azure AI Translator
• Google Cloud Translation vs Amazon Translate

Found a wrong rate or plan restriction? Report a correction with a current source.