If you are deciding whether to ship a multilingual product on DeepSeek, the practical question is not “does it speak French?” but “how well, on which model, at what cost, and with what failure modes?” This guide answers all four for the DeepSeek languages that matter to most teams in 2026, covering both natural languages and programming languages on the current generation, DeepSeek V4.

I have run V4-Pro and V4-Flash in production since the preview shipped on April 24, 2026, and ran V3, V3.2 and R1 before that across English, Mandarin, Spanish, French, German, Japanese and a handful of lower-resource languages. Below: the supported list, quality tiers I have observed, and the prompts and parameters that change the answer.

The short answer: what DeepSeek actually supports

DeepSeek does not publish a closed list of “officially supported” natural languages the way some translation APIs do. Instead, V4-Pro and V4-Flash are general-purpose multilingual models trained on a corpus dominated by English and Chinese, with meaningful coverage of the major European, East Asian and a growing set of South and South-East Asian languages. In day-to-day testing I see roughly three tiers:

• Tier 1 — production-grade: English, Simplified Chinese, Traditional Chinese.
• Tier 2 — strong, with light review: Spanish, French, German, Portuguese, Italian, Russian, Japanese, Korean, Dutch, Polish, Arabic, Turkish, Vietnamese, Indonesian, Hindi.
• Tier 3 — usable but verify: Swedish, Norwegian, Danish, Finnish, Czech, Greek, Hebrew, Thai, Malay, Tagalog, Ukrainian, Romanian, Hungarian, Bengali, Tamil, Urdu, Swahili.

Anything outside those tiers — particularly low-resource African and indigenous languages — works inconsistently. If your use case targets one of those, benchmark before you commit. For background on what V4 is and how it differs from earlier releases, see our overview of DeepSeek V4.

Why the model is bilingual-first

DeepSeek’s published technical reports through V3 described a training mix weighted toward English and Chinese, with multilingual data added to broaden coverage rather than to match dedicated translation systems. V3.1 and V3.2 narrowed the gap on European languages; V4 narrows it further but does not change the fundamental shape: English and Chinese are the languages the model was optimised against, and benchmarks are reported in those two languages by default.

Practically, this means three things:

1. English in, English out is the smoothest path. Reasoning quality, instruction-following and tool calling are most reliable.
2. Chinese is a near-peer for most tasks, including code commentary and structured output.
3. Other languages work — sometimes very well — but you should expect occasional code-switching into English on long generations, especially when thinking mode is on.

Natural language coverage at a glance

The table below reflects my own production testing across V4-Flash and V4-Pro between April 24 and April 25, 2026. “Quality” is a subjective rating for chat, summarisation and translation tasks against a native-speaker review. It is not a benchmark.

Programming languages

For code, DeepSeek inherits the lineage of DeepSeek Coder V2 and the V3-series coding work. V4-Pro is the stronger choice for agentic and repository-scale tasks; V4-Flash is the default for autocomplete-style and single-file work. Both ship the same language coverage. The tiers I rely on:

• First-class: Python, JavaScript, TypeScript, Java, C, C++, C#, Go, Rust, SQL, Bash, HTML, CSS.
• Strong: Kotlin, Swift, PHP, Ruby, Scala, R, MATLAB, Lua, Dart, Objective-C, Perl, Haskell, Elixir, Erlang.
• Usable with review: Solidity, F#, OCaml, Clojure, Julia, Fortran, COBOL, VHDL, Verilog, Assembly variants.

Fill-in-the-Middle (FIM) completion is available on V4 in non-thinking mode only — useful for editor integrations. Tool calling, JSON mode and streaming all work in either thinking or non-thinking mode. For deeper coding workflows, see DeepSeek for coding.

Choosing the right model for multilingual work

Both V4 tiers are open-weight Mixture-of-Experts models released under the MIT license. They share a 1,000,000-token context window with output up to 384,000 tokens and the same OpenAI- and Anthropic-compatible API surface. The differences that matter for language work:

Pricing is as of April 2026; check the DeepSeek API pricing page before sizing a budget. The legacy IDs deepseek-chat and deepseek-reasoner still work and currently route to deepseek-v4-flash, but they retire on July 24, 2026 at 15:59 UTC. Migrating is a one-line model= swap; the base_url does not change.

How thinking mode changes language behaviour

Thinking mode is a request parameter on either V4 model, not a separate model ID. You enable it with reasoning_effort="high" plus extra_body={"thinking": {"type": "enabled"}}, or push to reasoning_effort="max". The response then returns reasoning_content alongside the final content.

For multilingual work this matters in two ways. First, the reasoning trace is often produced in English even when the user prompt is in another language — this is normal and does not affect the final answer. Second, on Tier 3 languages I see slightly better translations with thinking on, at the cost of latency and tokens. For a quick chat reply in Spanish, leave it off; for a 3,000-word legal translation into Polish, turn it on.

Calling the API in another language

Chat requests hit POST /chat/completions, the OpenAI-compatible endpoint at https://api.deepseek.com. The API is stateless — you must resend the conversation history with every request. The web chat and mobile app, by contrast, maintain session history for the user. A minimal Python example for a Japanese translation request using the OpenAI SDK:

from openai import OpenAI

client = OpenAI(
    base_url="https://api.deepseek.com",
    api_key="YOUR_KEY",
)

resp = client.chat.completions.create(
    model="deepseek-v4-flash",
    messages=[
        {"role": "system", "content": "You are a professional EN-JA translator. Output only Japanese."},
        {"role": "user", "content": "Translate: 'The quarterly report is attached for your review.'"},
    ],
    temperature=1.3,
    max_tokens=512,
)
print(resp.choices[0].message.content)

Three parameter notes for language tasks. temperature follows DeepSeek’s official guidance: 0.0 for code, 1.0 for data analysis, 1.3 for general conversation and translation, 1.5 for creative writing. max_tokens should be set generously for non-Latin scripts, which can use more tokens per equivalent meaning. top_p is an alternative to temperature — I rarely change both at once.

Translation prompting that actually works

The single biggest quality lever is not the model tier — it is the prompt. Five patterns I rely on:

1. Specify variant. “Latin American Spanish, neutral register” beats “Spanish.”
2. Pin the audience. “For a B2B SaaS landing page aimed at Berlin-based engineering managers” sets register without you having to enumerate it.
3. Forbid code-switching. “Output only in Polish. Do not include English glosses.” removes a common Tier 3 failure mode.
4. Provide a glossary. A six-line “translate these terms exactly” block fixes brand names and domain vocabulary better than any temperature tweak.
5. Round-trip for high-stakes copy. Translate to target, then back to English with a fresh request, and compare. Cheap on Flash, revealing on tone.

For more on prompt structure, see DeepSeek prompt engineering, and for a translation-specific deep dive, DeepSeek for translation.

Cost example for a translation workload

A worked example on deepseek-v4-flash, sized for a localisation pipeline pushing 1,000,000 segments per month with a 2,000-token system prompt (cached), a 200-token source segment per call, and a 300-token translated output:

• Input, cache hit: 2,000 × 1,000,000 = 2,000,000,000 tokens × $0.028/M = $56.00
• Input, cache miss: 200 × 1,000,000 = 200,000,000 tokens × $0.14/M = $28.00
• Output: 300 × 1,000,000 = 300,000,000 tokens × $0.28/M = $84.00
• Total: $168.00

The same workload on deepseek-v4-pro costs $1,682.00 — roughly 10× — because Pro’s output rate is $3.48/M. Reserve Pro for the segments where reasoning quality moves the needle (legal, medical, literary). For most localisation, Flash is the right answer. If you want to model your own numbers, the DeepSeek pricing calculator handles the three-bucket math.

Known failure modes by language

• Code-switching to English mid-output: most common in Tier 3 languages and in thinking-mode responses. Fix with a “respond only in <language>” instruction in the system prompt.
• Simplified vs Traditional Chinese drift: specify “Traditional Chinese (Taiwan)” or “Traditional Chinese (Hong Kong)” explicitly.
• Honorific register slippage in Japanese/Korean: pin the register (“desu/masu form throughout”).
• RTL display issues: Arabic and Hebrew render correctly from the model; bugs are usually in your frontend, not the response.
• JSON mode with non-Latin keys: keep keys ASCII; values can be any language. JSON mode is designed to return valid JSON, not guaranteed — include the word “json” and a small example schema in your prompt, and set max_tokens high enough to avoid truncation.

Where DeepSeek fits versus dedicated alternatives

For volume translation between two specific high-resource languages — say, English ↔ German for a documentation pipeline — a dedicated translation API can still edge DeepSeek on raw fluency, especially on idioms. DeepSeek wins when the task is more than translation: summarising a meeting transcript in one language and producing a localised brief in another, drafting a multilingual support reply that cites the original ticket, or translating code comments while preserving semantics. For a side-by-side with a frontier general-purpose model, see DeepSeek vs Claude, and for the broader landscape, DeepSeek alternatives.

UI language for the chat and app

Separate question: what language is the chat interface itself in? The DeepSeek app and web chat default to the device or browser language for menus and system messages, with English and Chinese as the most polished surfaces. The model will respond in whichever language you write in, regardless of UI language.

Practical picks by scenario

• English-only product, occasional Spanish/French support tickets: V4-Flash, non-thinking, temperature 1.3.
• Bilingual zh-en knowledge base: V4-Flash, context caching enabled, system prompt fixes the output language per request.
• High-stakes legal or medical translation: V4-Pro with thinking on, native reviewer in the loop.
• Multilingual coding assistant: V4-Pro for repo-scale work, V4-Flash for editor autocomplete via FIM.
• Low-resource language research: benchmark first; consider DeepSeek limitations and have a fallback.

For a wider tour of the ecosystem, the DeepSeek beginner guides hub collects the prerequisites — accounts, keys, model picks — in one place.

Last verified: 2026-04-25. DeepSeek AI Guide is an independent resource and is not affiliated with DeepSeek or its parent company. Model IDs, pricing and API behaviour change; check the official DeepSeek documentation and pricing page before committing to a production decision.