multilingual-support changedbuild-with-claude/multilingual-support
Nearest release: v2.1.318, published 4 hours before this site recorded the change. Shown because the two are within 24 hours of each other. Nothing here says the release caused the edit.
Recorded here
Lines+26added
Lines−26removed
From line
33
where the diff opens
First seen
14 Aug 2026
this site's first read of the page
Recorded edits13to this page, all time
The whole hunk
from line 33, old and new numbered
/
from line 33
3333
3434## Overview
3535
36Claude demonstrates robust multilingual capabilities, with particularly strong performance in zero-shot tasks across languages. The model maintains consistent relative performance across both widely spoken and lower-resource languages, making it a reliable choice for multilingual applications.
36Claude demonstrates robust multilingual capabilities, with particularly strong performance in zero-shot tasks across languages.
3737
3838Claude is capable in many languages beyond those benchmarked in the following table. Test with any languages relevant to your specific use cases.
3939
4040## Performance data
4141
42The following table shows zero-shot chain-of-thought evaluation scores for Claude models across languages, expressed as a percentage relative to English performance (100%):
42The following table shows zero-shot scores for Claude models on [Global-MMLU](https://arxiv.org/abs/2412.03304), which extends the [MMLU (Massive Multitask Language Understanding)](https://en.wikipedia.org/wiki/MMLU) test set to 42 languages. Each score is a percentage of the same model’s English score:
4343
44| Language | Claude Sonnet 4.5 (deprecated)<sup>1</sup> | Claude Haiku 4.5<sup>1</sup> |
45| --------------------------------- | ------------------------------------------ | ---------------------------- |
46| English (baseline, fixed to 100%) | 100% | 100% |
47| Spanish | 98.2% | 96.4% |
48| Portuguese (Brazil) | 97.8% | 96.1% |
49| Italian | 97.9% | 96.0% |
50| French | 97.5% | 95.7% |
51| Indonesian | 97.3% | 94.2% |
52| German | 97.0% | 94.3% |
53| Arabic | 97.2% | 92.5% |
54| Chinese (Simplified) | 96.9% | 94.2% |
55| Korean | 96.7% | 93.3% |
56| Japanese | 96.8% | 93.5% |
57| Hindi | 96.7% | 92.4% |
58| Bengali | 95.4% | 90.4% |
59| Swahili | 91.1% | 78.3% |
60| Yoruba | 79.7% | 52.7% |
44| Language | Claude Fable 5.1 | Claude Opus 5.5 | Claude Sonnet 5.5 | Claude Haiku 5.5 |
45| -------------------- | ---------------- | --------------- | ----------------- | ---------------- |
46| English | 100% | 100% | 100% | 100% |
47| Spanish | 99.4% | 99.3% | 99.0% | 98.1% |
48| Portuguese (Brazil) | 99.0% | 99.2% | 98.7% | 98.1% |
49| Italian | 99.0% | 99.2% | 98.4% | 97.7% |
50| French | 99.0% | 99.3% | 98.7% | 97.7% |
51| Indonesian | 98.9% | 99.1% | 98.6% | 97.1% |
52| German | 99.1% | 99.1% | 98.5% | 97.4% |
53| Arabic | 98.7% | 99.3% | 98.3% | 96.6% |
54| Chinese (Simplified) | 98.8% | 99.3% | 98.5% | 97.7% |
55| Korean | 98.8% | 99.1% | 98.3% | 97.0% |
56| Japanese | 98.7% | 99.2% | 98.4% | 97.4% |
57| Hindi | 98.9% | 99.2% | 98.5% | 96.6% |
58| Bengali | 98.3% | 98.6% | 97.2% | 95.4% |
59| Swahili | 97.7% | 98.3% | 96.9% | 94.0% |
60| Yoruba | 95.1% | 96.3% | 93.6% | 87.0% |
6161
62<sup>1</sup> With [extended thinking](https://platform.claude.com/docs/en/build-with-claude/extended-thinking).
62All scores use [adaptive thinking](https://platform.claude.com/docs/en/build-with-claude/thinking) at the `max` [effort level](https://platform.claude.com/docs/en/build-with-claude/effort). Scores exclude questions that safety classifiers blocked or that went unanswered. For each score in the table, at most 1.7 percent of the questions are excluded.
6363
6464<Note>
65 These metrics are based on [MMLU (Massive Multitask Language Understanding)](https://en.wikipedia.org/wiki/MMLU) English test sets that were translated into 14 additional languages by professional human translators, as documented in [OpenAI’s simple-evals repository](https://github.com/openai/simple-evals/blob/main/multilingual_mmlu_benchmark_results.md). The use of human translators for this evaluation ensures high-quality translations, particularly important for languages with fewer digital resources.
65 Use the table to compare languages within a model, not to rank models. Differences of less than about 1 percentage point are within measurement noise.
6666</Note>
6767
6868***
from line 76
7676 curl https://api.anthropic.com/v1/messages \
7777 -H "content-type: application/json" \
7878 -H "x-api-key: $ANTHROPIC_API_KEY" \
79 -H "anthropic-version: 2023-06-01" \
79 -H "anthropic-version: 2026-10-21" \
8080 -d '{
8181 "model": "claude-opus-5-5",
8282 "max_tokens": 1024,
from line 230
230230 </Card>
231231
232232 <Card title="Customer support agent" icon="headset" href="https://platform.claude.com/docs/en/about-claude/use-case-guides/customer-support-chat">
233 Build a localized support chatbot using a language-constrained system prompt.
233 Build a customer support chatbot. One chatbot can serve customers in many languages.
234234 </Card>
235235
236236 <Card title="Models overview" icon="table" href="https://platform.claude.com/docs/en/models/overview">
from line 238
238238 </Card>
239239
240240 <Card title="Define success criteria and build evaluations" icon="scales" href="https://platform.claude.com/docs/en/test-and-evaluate/develop-tests">
241 Evaluate translation and localization quality before you ship.
241 Apply general evaluation methods to measure output quality in each language you support.
242242 </Card>
243243</CardGroup>
244244
No line in this hunk matches that.