The best translation providers for Japanese games (2026)

Picking a translation provider is the most consequential decision in the workflow. Engine compatibility decides whether a game can be translated at all; the provider decides whether the result is readable, in-character, and affordable. This is the guide to the best providers for translating Japanese games— RuneTranslate ships nine, from free no-key engines to premium LLMs to fully on-device models, and you can mix them per project.

On a paid plan you can also use provider routing to send short strings (item names, menu labels) to a cheap provider and longer dialogue to a premium one — more on that below.

The premium LLMs — best character voice

Large language models read the surrounding lines and keep register consistent (formal NPC vs colloquial party member). They are the best choice for dialogue-heavy games where character voice matters.

Anthropic Claude — the eroge / VN specialist

Best at: preserving tone, character voice, and onomatopoeia — the three things visual novels live and die by. Claude is more willing than GPT to render stylized or dialectal speech instead of normalizing it, and more willing to handle adult content in eroge / doujin work without over-sanitizing. RuneTranslate defaults to Claude Haiku 4.5 (cheap, ~$1/$5 per million tokens, strong on short strings) with Claude Sonnet as the quality option (~$3/$15). Pair it with a glossary and it is near-publication-quality on long VNs. Use for: visual novels, especially adult ones.

OpenAI GPT — context-aware all-rounder

Best at: context-aware translation with consistent register and good English slang. The default is gpt-4o-mini (cheap and fast — right for most games); gpt-4o is the quality option (about $2.50/$10 per million tokens) for a stronger read. Watch out: it can drift register across long runs and sometimes refuses adult content. Use for: visual novels with strong character voice where you don't need Claude specifically.

DeepSeek — LLM quality at MT prices

Best at: the same context-aware quality for a fraction of the cost. deepseek-v4-flash (the default) runs roughly $0.14/$0.28 per million tokens — a whole VN for cents — with deepseek-v4-pro for the hardest dialogue. It is the value pick when you want LLM quality without GPT or Claude prices; see our guide to translating with DeepSeek. Use for: long games on a budget.

OpenAI-Compatible — bring any model

Best at: flexibility. Point RuneTranslate at any OpenAI-compatible endpoint — OpenRouter, NanoGPT, and dozens of gateways — to use models RuneTranslate doesn't ship natively (Gemini, Llama, Qwen and more) with your own key and base URL. Use for: a specific model you already prefer.

The dedicated MT engines — fast and cheap

DeepL — the narrative workhorse

Best at: long-form Japanese narrative. DeepL's neural MT is tuned for JA ↔ EN and produces fluent, naturally-phrased English other MT underperforms on. Cost: a keyed free tier covers 500,000 characters/month; paid plans start around $7/month. Falls short on: stylized speech, slang, and cross-line context — every character ends up sounding the same. Use for: RPG Maker, Wolf RPG, kinetic novels — especially first drafts you'll hand-edit.

Free DeepL — no key, no cost

RuneTranslate can also reach DeepL's free anonymous endpoint with no API key, including a choice of DeepL's Classic or newer next-gen model. It is rate-limited (best for small or occasional jobs), but it is the same DeepL quality for $0. Use for: free-tier users who want DeepL-grade dialogue without signing up.

Google Translate — the free fallback

Best at: being free and fast for short isolated strings — item names, button labels, hai / iie options — with no key. Falls short on: dialogue, and it sometimes returns the source unchanged for onomatopoeia (RuneTranslate flags those as failed so you can retry on a smarter provider). Use for: the short-string lane in provider routing.

On-device — private and unlimited

Local model — Ollama / LM Studio

Best at: privacy and volume. Point RuneTranslate at a model running on your own PC through Ollama or LM Studio and nothing leaves your machine — no API key, no per-word cost, no content policy. Quality depends on the model you run, and it is as fast as your GPU. Use for: sensitive content, offline work, or grinding huge games at zero marginal cost.

Recommendation by game type

• JRPG (RPG Maker MV / MZ, Wolf RPG): DeepL or DeepSeek for dialogue, free Google for item names via routing.
• Visual novel (Kirikiri, TyranoBuilder, Ren’Py): Claude or DeepSeek for everything; GPT a close second; DeepL is fine if you'll hand-edit anyway.
• Eroge / adult VN: Claude or a local model — GPT may refuse content, DeepL and Google flatten the voice.
• Free-tier user, any genre: free Google for the first pass, then re-run the failed dialogue lines on free DeepL.

Provider routing makes the choice less binary

On a paid plan, RuneTranslate's provider routing lets you set a length threshold (default 12 characters) and send everything shorter to a cheap provider while keeping dialogue on your premium one. For a typical VN — where most of the unit count is short item / menu strings, not dialogue — this cuts your LLM bill by 60–80%. A common setup: DeepSeek or Claude on dialogue, free Google on the short-string lane, for roughly a dollar a game.

Quality is also about hand-editing

No provider produces a publication-ready translation on the first pass. RuneTranslate's editor exists for the hand-fix pass: the provider gets you ~90% of the way there, and you get the last 10%. On Supporter / Pro those hand-edits are cached in translation memory— the same Japanese line in a different game reuses your fix instantly, free. Over a few games this compounds into a private memory of game-specific terminology the machine providers consistently miss.

Download RuneTranslate to try any of these providers head-to-head on a real game. The free tier works with every provider — just modestly slower.