HFAI Cloud · catalogue
Every model,
one catalogue.
Eleven entries, one endpoint and one key. Here is what each row is.
Eleven rows, side by side.
Developer, licence and context window,
with a link to each model's page.
This is the catalogue rather than the price list: one row for each of the eleven entries the API serves, in the order the model picker presents them. A row records the name you see, the developer that built the model, the licence its card states and the context window the live catalogue returns for it. Each name links to that model's own page, which carries the request example and the dated sources behind these values. Prices change independently of any of this, so they live on the pricing page and are not repeated here.
Three rows show a dash. Gemma 4 is served as a family alias rather than one checkpoint, so no single card can be tied to it; the two entries that exist only as catalogue names have no public card at all. Leaving those cells empty is deliberate: a developer or licence guessed from a similar model would be a claim the sources do not support. The other eight rows name the project that built the model and the terms it ships under. Developer and licence values were read from each row's recorded source on ; Gemma 4's search is dated .
| Model | Developer | Licence | Context |
|---|---|---|---|
DeepSeek V4.1 Flashdeepseek-v4.1-flash | DeepSeek | MIT | 128K |
GPT-OSS 120B FREEgpt-oss:120b | OpenAI | Apache-2.0 | 128K |
GPT-OSS 20Bgpt-oss:20b | OpenAI | Apache-2.0 | 128K |
Gemma 4gemma4 | — | — | 128K |
GLM 5.3 Flashglm-5.3-flash | Z.ai | MIT | 128K |
MiniMax M3minimax-m3 | MiniMax | MiniMax Community | 128K |
DeepSeek V4 Prodeepseek-v4-pro | DeepSeek | MIT | 1M |
Nemotron 3 Ultranemotron-3-ultra | NVIDIA | OpenMDW 1.1 | 256K |
Kimi K2.6kimi-k2.6 | Moonshot AI | Modified MIT | 256K |
Neira H4neira-h4 | — | — | 256K |
GLM 5.3 Abliteratedglm-5.3:abliterated | — | — | 1M |
Only GPT-OSS 120B is marked free: it is the row the anonymous tier covers, and the pill beside its name says so. Everything else draws on prepaid credits at the published rate. The context column is the ceiling a single request can carry — 128K for six rows, 256K for three and 1M for two — and it does not imply that every client is configured to send that much. The identifier under each name is the exact string to place in the body of a chat request; the same key and the same endpoint serve all eleven.
What each row costs is on the pricing page, and the live catalogue itself is published at https://api.heartflowai.net/v1/models. No speed, quality or benchmark figure is claimed on this page.
Pick a row.
Send its identifier.
Every entry above answers on the same endpoint and draws on the same balance, so comparing them costs nothing until you call one.