Zhipu AI English: accessing the platform from outside China

The English UI toggle

Both the Z.ai chat surface and the BigModel developer console have an English toggle that persists once set for signed-in accounts.

The Z.ai chat surface and the BigModel developer console both default to Chinese for new visitors. This is the correct default for the platform's primary user base, but it creates a moment of confusion for outside teams who land for the first time. The zhipu ai english toggle is present in both surfaces, though its location differs slightly between them.

On the chat surface, the language toggle is typically in the account settings dropdown accessible from the user avatar in the top-right corner. On the BigModel console, the language selector appears in both the account settings page and as a globe icon in the top navigation bar on most pages. On either surface, selecting English once will persist the preference for signed-in sessions — a subsequent visit will load in English without requiring the toggle again.

For developers using the API without a browser session, the language of the interface is irrelevant — the API itself is language-agnostic and the model output language is controlled by the prompt language and system message, not by the console UI setting.

English-tuned model behaviour

The GLM family was trained with strong English and Mandarin coverage; flagship and mid-tier models produce professional-quality English output without additional prompting.

The GLM family is genuinely multilingual rather than translated-multilingual. The training corpus for the mid-tier and flagship generations included substantial English data alongside Chinese, and the instruction-tuning phase used English-language instruction datasets in addition to Chinese ones. The practical result is that a team sending English-language prompts to a mid-tier or flagship GLM model will receive English outputs that read naturally for professional use cases — document summarisation, code explanation, technical writing assistance — without needing to specify output language explicitly in the system message.

The small-tier variants show more variation in English quality. For workloads that require a specific tone, formal register, or domain-specific vocabulary, a brief instruction in the system message — "Respond in formal English" — tightens the output reliably at all tiers. For workloads that mix Chinese and English in the same document (cross-lingual summarisation, bilingual email drafting), the flagship tier handles the switch natively in a way that smaller models occasionally stumble on.

Zhipu AI platform features — default language, English availability, and notes

Feature	Default language	English available	Notes
Chat surface UI	Chinese	Yes — toggle in account settings	Persists for signed-in sessions; reverts on logout for unregistered visitors
BigModel console UI	Chinese	Yes — globe icon in nav or account settings	A small number of settings sub-pages not yet fully localised
Model output (mid/flagship tier)	Prompt-language dependent	Yes — strong native English output	No additional prompting needed for English; formal register instruction helps for professional tone
API documentation	Chinese (primary)	Yes — English docs available	English docs occasionally trail Chinese docs by a few days on new API features
Billing emails and receipts	Chinese	Partial — amounts in CNY; some template strings remain Chinese	Platform team tracking full localisation of billing communications

International account registration

The BigModel registration flow accepts international phone numbers from most markets; country selection is available at the SMS verification step.

Registering a BigModel account from outside China requires an email address and a phone number capable of receiving an SMS. The registration page includes a country code selector at the phone number entry step. Most numbers from Western Europe, North America, Australia, New Zealand, and major Asia-Pacific markets (Japan, South Korea, Singapore, India) have been reported to work without additional steps. A small number of markets have sporadic delivery issues with the SMS gateway; in those cases, retrying after a few minutes or using a different number resolves the issue in most reports.

Once registration is complete, the account tier is the same regardless of the country of registration. There are no geo-restricted model tiers or feature restrictions for outside-China accounts at the standard plan level. Enterprise accounts with bespoke SLA requirements may involve separate onboarding, but for standard API and chat access the registration path is the same globally.

Non-CNY payment

International credit cards work through the BigModel payment gateway; billing is denominated in CNY with gateway conversion for non-Chinese cards.

The BigModel platform accepts Visa, Mastercard, and similar international credit-card networks. The billing cycle is monthly, and the invoice amount is stated in CNY with a gateway conversion applied at the prevailing rate for international cards. For teams that need a predictable hard-currency spend, the recommended approach is to pre-load a CNY credit balance in an amount equivalent to the expected monthly spend. The dashboard shows the CNY balance in real time, and the spend deducts from the balance rather than triggering a monthly gateway conversion event.

The platform does not currently offer contractual invoicing in USD, EUR, GBP, or other currencies for standard accounts. Enterprise accounts can discuss bespoke billing terms with the platform team. For most outside-China teams at the small-to-medium scale, the gateway conversion path with a pre-loaded balance covers the practical requirement. Guidance from NIST on AI procurement planning is useful background for teams formalising budget approval for ongoing API spend.

"The zhipu ai english toggle took less than thirty seconds to find. Our team in Annapolis ran the full onboarding flow — registration, API key, first English-language test call — in an afternoon. The billing setup with a pre-loaded CNY balance matched our procurement constraints cleanly."