Words, images, film.One API.
Language, image and video models behind one key and one base URL. The real models, straight from the labs that build them, billed transparently per request at a published price for every model.
Sign up with Gmail and get $0.39 (10,000 VND) to try the API right away. Start your free trial
- language models
- 29
- image models
- 13
- video models
- 7
The labs behind it
15 seconds,one request.
The film that opens this page — picture and sound — came from a single call to our video endpoint. No editing suite, no stock footage.
Three kinds of model.One account.
Chat, reasoning and coding models sit next to image and video generation — same key, same balance, same SDK. Add a modality without adding a vendor.
Language models
Leading chat, reasoning and coding models, called through the OpenAI and Anthropic formats your tools already speak — including Claude Code, Codex and other coding agents.
Live
- qwen4-max
- gemini-4-hydrogen
- gemini-4-carbon
- hy4
- +25 models
POST /v1/chat/completionsRead the docs
Image generation
Studio-quality images from a single sentence, edits that keep the composition, several frames per request, up to 2K.
Live
- gemini-nano-banana-2.1
- grok-imagine-image-2.0
- seedream-5.0-pro
- gemini-3.1-flash-lite-image
- +9 models
POST /v1/images/generationsRead the docsVideo generation
Text or images into 1 to 15 second videos, then edit and extend the clips you've made.
Live
- grok-imagine-video-1.5-lite
- seedance-2.5
- minimax-h3
- grok-imagine-video-1.5
- +3 models
POST /v1/videosRead the docsFor anyone who'sbeen burned before.
Renamed models, quietly degraded quality, keys that die mid-job, invoices that never match the dashboard. If that sounds familiar, this is for you.
Call a model, get that model
Every id is the genuine model from the lab that built it. No knock-offs borrowing a famous name, no silent swap to something cheaper.
Public, readable pricing
Each model's price is printed right on its row in the catalog, and every request is charged for exactly what it used.
Usage down to the request
The console breaks spend down by day, model and key, with every recent request and its status — so you can check every number we bill.
Steady under load
Limits wide enough for heavy traffic and capacity that stays ahead of demand, so a render batch or an agent session runs to the end.
Your data stays yours
We don't keep prompts, input images or outputs — only what it takes to bill a request.
Real people on the other end
Questions go straight to the team that runs the platform, not into a ticket queue.
What changes when you switch
Elsewhere
- A famous model name on a cheaper model
- Invoices and dashboards that never match
- Keys throttled or dead in the middle of a job
- Support is a ticket nobody answers
At Framewharf
- The genuine model the id names, from the lab that built it
- Public prices and per-request usage
- Limits wide enough for heavy traffic
- Message the operators directly on Telegram
Pay as you go,or unlimited Grok.
Run any model and pay only for the requests you make, or buy a 7-day Grok pass by concurrency for as many Grok images and videos as you like.
Pay as you go
Flexible
Per request
published price for every model
- Every language, image and video model
- Top up any amount, credit never expires
- Per-request usage in the console
- $0.39 (10,000 VND) free trial when you sign up with Gmail
Grok pass
1× concurrency
$17.13 (441,000 VND)
/ 7 days
- Unlimited Grok images and videos
- 1 requests in parallel
- Valid for 7 days
Grok pass
2× concurrency
$34.26 (882,000 VND)
/ 7 days
- Unlimited Grok images and videos
- 2 requests in parallel
- Valid for 7 days
Grok pass
3× concurrency
$51.38 (1,323,000 VND)
/ 7 days
- Unlimited Grok images and videos
- 3 requests in parallel
- Valid for 7 days
Grok pass · custom
Need more concurrency?
Pick anywhere from 1 to 50.
Each parallel request costs $17.13 (441,000 VND) for 7 days, with unlimited Grok images and videos — 10 in parallel is $171.28 (4,410,000 VND). Add more mid-period and pay only for the time left.
Choose concurrency

















A reel madeentirely of the real thing.
Every clip here came out of the API — animated from a single still, or written from a single sentence. Hover a clip to see the model that made it.
Integrate once,call every model.
OpenAI-format endpoints — plus the Anthropic format for Claude — so the SDK your team already uses works as is.
- 1Sign in with Gmail
Your first sign-in creates an account with $0.39 (10,000 VND) to try things out; top up by bank transfer whenever you need more.
- 2Create a key
Right in the console — one key per product, each with its own budget if you like.
- 3Point your SDK at Framewharf
Swap the base URL and key — the rest of your code stays the same.
from openai import OpenAI
client = OpenAI(
base_url="https://api.framewharf.com/v1",
api_key="sk-your_api_key",
)
reply = client.chat.completions.create(
model="qwen4-max",
messages=[{"role": "user", "content": "Write a short tagline for a coffee shop."}],
)
print(reply.choices[0].message.content)Every model,one key.
Language, image and video models on the same API. Filter by type, search by name or id, and see each model's price and capabilities right on its row.
49 models available
The firstquestions people ask.
What you'll want to know before your first request.
Yes. Every id is served by exactly the model it names, from the lab that built it. We never answer a request for one model with another, and a model we can't serve properly is removed from the catalog, not substituted.
Get started with Framewharf
One key.Every model.
Sign in with Gmail for free trial credit, create a key, and call language, image or video models in minutes. Need help choosing a model? Message us.


