IN ENGLISH · CHANGELOG

Changelog

What changed for every account, by date: the API, billing, documentation and the product. Each item is listed on the day it was available to everybody.

Last updated:

  • Billing

    Service credit: the rate that decides a credit counts only the failures that were ours (our gateway's, servers' and routes' errors and cuts by our own timers, or a problem with our own account at the vendor, such as our credit there running out). A vendor's own outage is neither a failure nor in the denominator. This holds from Azar 1405, the commitment's first month. The status page shows the same rate as before, and the service credit page shows both rates.

  • Billing

    The service credit's tiers changed before its first month: under a 99.5% monthly success rate the credit is 10%; under 98%, 15%; under 95%, 20%; and under 90%, 25% of what your wallet paid that vendor in the month. 25% is the most a month pays. The tiers hold from Azar 1405, and our team still reviews each month's figures before anything is paid (service credit).

  • API

    Per-key controls: an API key can be held to a list of models, a list of IP addresses and a scope (model calls only, or reading reports only), and given a rate limit of its own below the account's; a key can be rotated with a grace period for the old secret. A key with no controls works as before, and a request a control refuses answers 403 and is not billed (key controls).

  • Billing

    A service credit: when a vendor family's monthly success rate is under 99.5%, 10, 25 or 50 percent (by the rate) of what your wallet paid for that vendor in the month is credited back. Months are Persian (Jalali) calendar months, covered from Azar 1405; our team reviews each month's figures before anything is paid (service credit).

  • API

    OpenAI Batch is open to every account: upload OpenAI chat and embedding requests as one JSONL file with POST /v1/files and create a batch with POST /v1/batches; results are ready within 24 hours, and input and output are billed at 50% of the usual price, with the same markup. A batch is paid from the wallet only, and only completed requests are billed (OpenAI Batch).

  • Product

    The chat shows an estimate in Toman before a message is sent; it can be hidden in the settings.

Release 19 and 20

  • API

    The default limit of an API key went from 300 to 1,000 requests a minute. Every API response carries the headers X-RateLimit-Limit, X-RateLimit-Remaining and X-RateLimit-Reset, and a higher limit for the account can be requested from the keys page.

  • API

    When a model's vendor is busy, the 429 on the OpenAI-shaped routes has code upstream_capacity; rate_limit_exceeded now means the key's own limit only. A vendor's own rate-limit headers are no longer forwarded.

  • API

    Every API response carries X-Request-Id: 1xAi's own id of the call, which is also the request_id of its usage row (it used to be the vendor's).

  • Billing

    The non-streamed answer of a billed call states its amount in Toman in the X-1xai-Cost-Toman header; the amount of a streamed call is in the usage report under the same request id.

  • API

    The usage report: GET /v1/usage?request_id=… finds one call exactly, and GET /v1/usage/export (and a download button on the dashboard's call log) returns a date range as CSV. from and to take a plain date (a Tehran day) on the usage routes and on the wallet's transactions, and a value that cannot be read answers 400 instead of the whole list. The q search stops after a few seconds with 503 and code usage_search_timeout.

  • API

    GET /v1/keys/current returns the key the request was sent with. Each model of GET /v1/models has a type field, and chat and embedding models have context_window and max_output_tokens when known.

  • API

    The 401 answers for want of a usable key carry a code: missing_api_key, invalid_api_key and account_disabled. An unknown path under /v1/usage, /v1/wallet or /v1/keys answers 404 with code path_not_found.

  • API

    The other objects a vendor keeps for a call (Gemini files, operations and interactions, videos) are also reachable only with keys of the account that created them, and only with keys of the same company workspace when a workspace key created them; any other key gets 404. A vendor's own listing of such objects is not served through 1xAi.

  • API

    Anthropic Message Batches are documented, and a batch's results_url now points at 1xAi, so the Anthropic SDKs download the results through the gateway with your 1xAi key.

  • Product

    The status page shows response times per vendor and model from real traffic (first byte, total time and generation speed), and a benchmark page explains the method and has a script to measure for yourself.

  • Product

    The student plan has a public page, in Persian (student plan): its price, what it includes and how a university's students unlock it.

  • Docs

    The documentation was split into one page per topic, in Persian and English; the OpenAPI file was rewritten, and this changelog was added.

Hotfix

  • API

    Anthropic message batches are reachable only with keys of the account that created them, and only with keys of the same company workspace when a workspace key created them; any other key gets 404. The list of batches is not served through 1xAi.

  • API

    On the account API a company workspace key reads its own workspace only: that workspace's calls and keys, and the spendable balance. The wallet ledger is on the company dashboard, and GET /v1/wallet/transactions answers 403 to such a key. A personal key reads the calls and keys outside company workspaces. Through an API key, logged request and response text is returned only for calls made with an API key.

Release 12

  • Product

    The site's pages and the chat became lighter and open faster.

Release 10, 10.1 and 11

  • API

    The model codex-auto-review was added to the model list and the pricing page, marked as a special offer.

  • API

    Two new error codes for special-offer models: 503 with code offer_unavailable and a Retry-After header, and 400 with code offer_request_refused.

  • Product

    The tutor answers with a faster model by default and offers a stronger one.

  • Product

    Fixed: navigation could freeze after leaving a conversation.

  • Product

    The pricing page became lighter.

Release 8 and 9

  • API

    Parameters that OpenAI reasoning models refuse (temperature, top_p and the repetition penalties) are removed from a request instead of returning a 400, and max_tokens becomes max_completion_tokens for those models.

  • Billing

    A call that is cut after delivering no answer is not charged.

  • API

    The Gemini transcription model works on /v1/audio/transcriptions with OpenAI's form.

  • API

    A problem with our own account at a vendor no longer reaches a customer as a 401 or 402; the answer is 503 with code upstream_unavailable.

  • API

    An upload over the size limit answers 413.

  • Product

    The chat composer was redesigned, with a "+" button and a voice-mode screen.

  • Product

    When a model does not answer, the chat tries another model automatically.

  • Billing

    Empty or failed chat turns are refunded.

Release 7.1 and 7.2

  • Product

    The privacy policy, edition 2.3, was published in Persian and English.

  • Product

    Fixed: editing a message emptied it, and a few pages were wider than a phone's screen.

Release 7

  • Product

    The wallet notice closes with one tap, and form fields became easier to read.

Release 6

  • Product

    The whole site was re-toned: a light ground by day and lapis by night.

Release 4 and 5

  • Docs

    Pages for the Claude models and a Claude Code guide were added.

  • Product

    A page of voice samples and new read-aloud voices were added.

  • Product

    New icons and illustrations across the site, and a new 404 page.

  • Billing

    A low-balance alert: you get an email when the balance falls under a threshold you set.

  • API

    Realtime voice with the Gemini Live models was added at wss://1xai.ir/rt/gemini.

  • API

    The vendor retired gpt-4.1-nano; its successor is gpt-5-nano.

  • Billing

    The markup became per model: 15% on flagship models, 30% on the small low-cost models and 20% on the rest; before that it was 20% on every model.

  • Billing

    The card-to-card top-up bonus became 10%.

The details of each API change are in the documentation; the service's current state is on the status page.

1xAi is an independent gateway. It is not affiliated with, endorsed by, or a reseller for OpenAI, Anthropic, Google, DeepSeek, xAI, Moonshot AI, Z.ai, Alibaba Qwen, MiniMax or Tencent Hunyuan. Requests are relayed upstream rather than answered by models of our own, so each provider's own usage policies apply to what their models will and will not do, and their model availability is theirs to change. Current reachability is published on the status page.