LatviaTechnology and telecom2025 – ongoing

Lokalise

AI translation suggestions in the Expert Editor

Lokalise translators accepted 82.6% of Claude's translation suggestions

WorkingVendor case studyVendor ↗

Problem

Lokalise sells a localization platform to software teams that publish content in many languages. Customers wanted translations that needed little post-editing, and no single model was best for every language pair.

Approach

Lokalise built an AI orchestration layer that compares output from several language models and picks one by language pair, content type and context. Claude 3.5 Sonnet is one of five engines. It is the default for several high-impact language pairs and handles longer segments such as product UI text and in-app messages. Translators see its suggestions in the Expert Editor and accept, edit or reject them. In an A/B test against GPT-4o, Claude had the higher acceptance rate.

Results

82.6%acceptance rate for AI translation suggestions made with ClaudeVendor case study[1]

80%cost savings compared with traditional translation methodsVendor case study[1]

over 80%share of AI translations that need no post-editingVendor case study[1]

Takeaway

Routing each language pair to the model that tested best gave Lokalise a measurable acceptance rate to track. The company plans to add retrieval of customer style and context, and to use Claude for quality scoring.

Sources

Lokalise improves translation quality with Claude
Anthropic12 Dec 2025Vendor publication

Source summaries

[1]Anthropic12 Dec 2025

Anthropic's customer story describes how Lokalise uses Claude 3.5 Sonnet as one of five engines in its AI translation orchestration. It reports an 82.6% acceptance rate for suggestions, 80% cost savings against traditional translation and over 80% of AI translations needing no post-editing. The page carries no date; the Internet Archive first saved it on December 12, 2025.

Full details are in the original sources.

Added 4 Nov 2026 · Updated 4 Nov 2026