AI Translation & Localization Startups · Google

Gemini

Google's LLM, used as an instruction-following translation engine via API.

★ Top Pick — AI Translation & Localization Startups. Highest measured score in this category: 99 / 100 (Grade A) in the Traditional Chinese Quality Benchmark v1.0, run 2026-11-21. See the run evidence →
99
Grade A · / 100
Traditional Chinese
Quality Benchmark v1.0
Script purity (simplified-character residue)100.0 / 100
Wording (mainland-term density)97.4 / 100
Terminology (preferred-term consistency)100.0 / 100

Measured 2026-11-21 · Gemini API, free tier, model gemini-3.5-flash-lite. All AI engines in this run received one shared neutral prompt ('Translate into Traditional Chinese as used in Taiwan' plus fixed formatting instructions); MT engines received the source text directly. Run 2026-11-21, 24/24 segments; raw outputs and prompt archived in the benchmark run. Raw outputs are published in the benchmark run archive. Method: how we test.

Conditions disclosure: AI/LLM engines received the corpus with one shared neutral prompt ('Translate into Traditional Chinese as used in Taiwan' plus fixed formatting instructions). Machine-translation engines received the source text directly, with no prompt. The groups were measured under different conditions; this disclosure travels with the scores wherever they appear.

Gemini is Google's large language model family, accessible through Google AI Studio and the Gemini API. Used as a translation engine, it takes natural-language instructions — target variety, glossary and tone — in the same call as the source text.

Strengths — from evidence & published facts

  • Zero simplified-character residue (0.00% of 381 CJK characters) and a perfect terminology record: 34 preferred renderings, 0 mainland variants
  • Only one wording flag in 24 segments — 手機號 in S3, a rule-list 'worth checking' flag (the fuller 手機號碼 is the more idiomatic form), not a script error

Watch-outs

  • This score belongs to the disclosed AI-group condition: an explicit prompt asked for Traditional Chinese as used in Taiwan. Machine-translation engines in the same run received no prompt — the conditions differ by design and are disclosed wherever scores appear
  • In two segments it paraphrased around the trap terms (M2 'quality' → 高音質; U5 → 高畫質); paraphrases are scored as 'other' — neither rewarded nor penalized in the terminology metric

Best for

Teams already using LLM APIs who want translation quality that follows instructions (glossaries, tone) without a separate MT contract.

Pricing (checked 2026-11-21)

Token-based API pricing with a free tier. Source: the vendor's official pricing page, checked 2026-11-21.

PlanPriceNotes
API free tierFreeRate-limited; the tier used for our benchmark run
API paid tiersToken-basedPer-token pricing on Google's official pricing page

Free tier: Free tier available through Google AI Studio / Gemini API · Trial: Free tier available

Visit the official site → Official pricing page

Score history (append-only)

Run dateBenchmarkScoreGrade
2026-11-21Traditional Chinese Quality Benchmark v1.099A

History is never rewritten: when a tool is re-tested, the new run is added and the old one stays.

Related in AI Translation & Localization Startups

  • Groq — High-speed LLM inference API — openai/gpt-oss-120b, benchmarked here as a translation engine.
  • OpenRouter — A single API routing to many LLMs, including free-tier models.
  • ElevenLabs Dubbing — Automatically dubs video and audio into other languages while preserving the original speaker's voice characteristics.
  • HeyGen — AI video platform that translates existing videos and re-syncs lip movement to the new language.
  • Rask AI — Video/audio localization platform that transcribes, translates and re-dubs with voice cloning.

Need this done by a human instead?

Tools route around quality problems; specialists remove them. HankLoc localizes websites, marketing, subtitles and games into Traditional Chinese — with a free one-page trial.

Localization services → Get a quote

Get the full benchmark report — and every future score — by email

The public pages show the headline scores. The email edition carries the complete dataset behind them:

  • The full per-segment benchmark dataset for every run (all 24 segments, every engine, every flag)
  • New-tool test alerts — the day a pending tool gets its first measured score
  • Quarterly re-test results, including score changes and what caused them

Get the benchmark report →

Free. One list, no spam, unsubscribe anytime. You will receive a confirmation email first.