SLA-backed.
Failures get refunded

ZetaAPI.AI is an AI API gateway for indie developers and small R&D teams. We expose API endpoints fully compatible with OpenAI, Anthropic and Gemini, and — without changing the interface — use dual-channel automatic failover, pay-as-you-go billing and a public compensation policy to make mainstream models more stable than the official endpoints and cheaper than a direct connection.

ChatGPT, Claude, Gemini, DeepSeek, Qwen, Kimi and other mainstream closed- and open-source models. The full model list and current price multipliers are on the pricing page.

Every model is pay-as-you-go; token-billed models have multipliers in the 0.3x–0.4x range (i.e. you pay 30%–40% of list price), while per-call models (e.g. image generation) are listed on the pricing page. The exact multiplier depends on upstream channel negotiations and is published transparently on the pricing page. Prompt Cache hit rates match official (cache reads are 10%–50% of the input price).

Three steps: ① sign in to the console with Google / GitHub / email → ② top up from $1 → ③ in Cursor / Claude Code / the OpenAI SDK, set the base URL to https://api.staging.cosmos-router.com/v1 and paste your new API key. About 30 seconds — see the quickstart.

We monitor availability for every model on every channel. When it stays below 95% for 5 consecutive minutes, compensation = your actual spend in that window × (1 − actual availability) × 2, credited to your balance. Currently claim-based: contact support to file, and we settle by the published algorithm after manual review (full automation is rolling out). The algorithm is published on the pricing page.