Price terms as published

Each line is the vendor’s own wording, split from the price capture checked on 2026-09-08. Nothing here is rounded, converted or re-worded — the full sentences stay in the review below. Lines that carry only a price continue the plan named on the line above, in the vendor’s own order.

3 price terms for Agnes AI
#Plan / termChecked
1¥0/M tokens (text)2026-09-08
2¥0/1K images2026-09-08
3¥0/sec video — all limited-time promo prices against published list rates2026-09-08

One-liner: Agnes AI (agnes-ai.cn) is a multimodal model gateway — text, image, and video generation behind one OpenAI-compatible API — that launched unlimited free pricing in June 2026 and currently bills ¥0 on its flash-tier models across all three modalities.

Verified facts (checked 2026-09-08, official docs)

Attribute Value Source
Base URL https://api.agnes-ai.cn/v1, OpenAI-style official docs overview, same day
Text: agnes-2.5-flash ¥0 / M input, ¥0 / M output — list prices ¥0.20 / ¥1.00 struck through official pricing page, same day
Text: agnes-2.0-flash Also ¥0, but flagged deprecated on the pricing page official pricing page, same day
Text: agnes-2.5-pro ¥3.00 / M in, ¥6.00 / M out (¥0.30 cache-hit) — not free official pricing page, same day
Text: agnes-2.5-pro-beta ¥0.70 / M in, ¥2.10 / M out — discounted, not zero official pricing page, same day
Images (agnes-image-2.0/2.1/2.5-flash) ¥0 per 1K output images (list: ¥70-160 per 1,000 images) official pricing page, same day
Video agnes-video-v2.0 ¥0/sec (list ¥0.035); agnes-video-2.5-flash ¥0/sec (list ¥0.15); agnes-video-2.5 is paid ¥0.15–0.35/sec official pricing page, same day
Endpoints Chat Completions / Responses / Messages compatibility claimed official docs model page
Signup API key self-serve at platform.agnes-ai.cn, no card mentioned official docs; secondary tester accounts agree

What the “indefinitely free” claim actually means

Third-party coverage is unusually consistent that this launch is real: when Agnes flipped its APIs to free on 2026-06-01, write-ups from QbitAI, Paicoding, CSDN and HKT Lab all independently confirmed working keys without a credit card, and paicoding cites vendor-reported weekly usage above 3T tokens. The company presents itself as Singapore-based (Sapiens AI); founder interviews circulated in Chinese tech media.

But read the pricing page the way Google reads it: every ¥0 sits next to a struck-through list price, and the page labels the ¥0 column as the “current (promotional) price” That is a revert signal written into the contract. The pro tier at ¥3/¥6 per million tokens is what Agnes charges when the promo ends on flash. Nothing on the official page states an end date; nothing on it promises the promo is permanent either.

Gotchas the marketing posts skip

  • agnes-2.0-flash is officially deprecated but still priced at ¥0. Migration is model-name-only; don’t pin the old ID in agent configs as a fallback.
  • “512K context / 20 RPM peak” is vendor-self-reported. As of today the official docs carry no standalone rate-limit page — searches of the docs return the overview. If your workflow needs throughput, you cannot verify the ceiling from the vendor.
  • The video tier split is easy to miss. agnes-video-2.5 (non-flash) costs money per second in every resolution; only -flash variants and v2.0 are at ¥0. Posts saying “video generation free” usually tested the flash line.
  • Single company, unaudited. Compared with AMD’s Token Factory (hardware vendor, published rate-limit docs) or Cline API (attached to a 67K-star open-source project), Agnes has no open-source footprint and no enterprise backer we can verify. Free-tier churn risk is highest here.

The pattern

Agnes is the model-vendor funnel shape in 2026’s free-API zoo: hardware vendors (AMD) bind developers to GPUs, tooling vendors (Cline) bind users to IDEs, and a small model lab binds users to its own weights — betting the promo converts to paid pro traffic before it collapses. When you evaluate “free API” channels, the durable question is which column reverts first.

Sources