Release notesRelease notes

Release notes

Changes to the Syntic API, the Syntic Console and the model catalog, newest first. Entries describe what changed for you as a developer: new behavior, new error codes, and anything you may need to act on.

For changes to the Syntic Code command-line tool, see the Syntic Code release notes.

Entries marked Action may require a change in your code or settings.

October 2026

October 12, 2026

October 4, 2026

  • API: the account and organization behind each credential are now checked on every request, for API keys and Syntic Code sign-in tokens alike. New responses:
    • 403 account_disabled when the key owner’s account is disabled;
    • 403 org_unavailable when the key’s organization is suspended or closed;
    • 503 authorization_unavailable when this check cannot be completed; retry with backoff.
  • A request that is already streaming is not interrupted by this check.
  • Action: add account_disabled and org_unavailable to the errors your code treats as “do not retry”, and authorization_unavailable to the retryable ones. See Errors.

October 3, 2026

  • Models: each model’s minimum plan now applies to API requests, as it already did in Syntic chat. A request for a model your plan does not include is answered by the default model, Amara S1.2, and the response carries an x-syntic-model-downgraded header naming the model that answered. See Models overview.
  • Action: if you request a model other than the default, log the x-syntic-model-downgraded header so a plan change cannot silently switch your model.
  • Billing: auto-reload only charges a payment method saved for the organization that owns the credit balance.
  • Billing: for organizations on a comped or contract plan, unmetered access now applies only to API keys that belong to that organization.

August 2026

August 6, 2026

  • API keys: an API key can carry an allowlist of web origins. When the list is set, a request must come from one of those origins (checked against the Origin header, then Referer), otherwise it is refused with 403 origin_not_allowed. Requests with no origin are refused when an allowlist is set. See Managing API keys.
  • Billing: one monthly dollar cap now bounds the total of all automatic card charges for an organization, including usage-credit auto-reload. Setting the cap to zero stops automatic charges altogether.

July 2026

July 18, 2026

  • Zero data retention: organizations in zero-data-retention mode can no longer reach models that need standard retention. Such requests return 400 model_unavailable before anything is billed. See Models overview.
  • Spend limits: administrators can set monthly credit spend limits at three levels: organization, group and individual member. A member or group at its limit receives 402 quota_exhausted with limit: "spend_limit". See Billing & credits.

July 16, 2026

  • Pricing: requests paid from usage credits are charged at per-model rates, with input, cached input and output priced separately. See Pricing.
  • Billing: before a credit-paid request runs, the balance must cover a worst-case estimate of its cost (based on max_tokens, or 4,096 when unset). A balance below that estimate returns 402 quota_exhausted with limit: "no_credits".
  • Action: set max_tokens on requests so small balances can still serve short requests.
  • Billing: if a client disconnects from a stream before the final usage figures arrive, the tokens produced so far are billed on an estimate rather than not at all.
  • Free plan: the Free plan’s token allowance is shared across all free organizations owned by the same person.

July 13, 2026

  • Usage windows: the weekly window is set to 60% and the 5-hour session window to 25% of the plan’s monthly allowance. The monthly allowance remains the main limit.
  • API: when usage cannot be verified, requests are refused with 402 quota_exhausted and limit: "unavailable" instead of being served unmetered. Retry with backoff.

July 11, 2026

  • Usage limits and credits: plan allowances are enforced over three windows (monthly, weekly, 5-hour session). When a window is full, requests continue on usage credits if credits are turned on, within the organization’s monthly spend limit.
  • New error body: 402 quota_exhausted now includes limit (monthly, session, weekly, spend_limit or no_credits), resets_at, upgrade_url and add_credit_url. See Handle usage limits.
  • Response header: x-syntic-billing-source reports whether a request was covered by the plan (subscription) or by usage credits (api_credit).

May 2026

May 24, 2026

  • Rate limits: each API key may send up to 600 requests a minute (per user for Syntic Code sign-in tokens), measured over a sliding 60-second window. Over the limit, the API answers 429 with Retry-After and X-RateLimit-* headers. See Limits.

May 23, 2026

  • Console: a new API Keys tab in Settings lets administrators and developers create and revoke keys. Key creation and revocation are recorded in the audit log.
  • API: request bodies are limited to about 2 MB (413 payload_too_large) and to 500 messages (413 too_many_messages).

May 18, 2026

  • Launch: the Syntic API opens with POST /v1/chat/completions at https://app.syntic.ai/v1, usable from OpenAI-compatible client libraries. Streaming is on by default.
  • Authentication: static API keys for third-party applications, sent as Authorization: Bearer <key>.
  • Models: the Amara models are available by their catalog names, with Amara S1.2 as the default.
Preview roadmap
Was this page helpful?