With 13 chat models next to image, video, speech and OCR, deAPI covers every step of a request: read it, decide, generate and check the result. The models sit behind an OpenAI-compatible endpoint and use the same key and credits.
$5 in free credits to start. No credit card required.
chat models, one endpoint
13
tokens of context on 8 models
1M
models read images as well as text
11
What a Chat Model Adds to deAPI
deAPI already transcribes, reads, generates and describes. A chat model reads those results, makes the call your rules ask for and writes the next request. Your code sends each request, so every step is logged and billed on its own.
Trust and safety
Screen partner requests before generation
Send the partner's prompt and reference image to a vision model with a JSON schema. It returns safe, the broken categories and a reason, so you reject a request before it reaches an image or video model.
Output review
Check generated media before delivery
Pass the result_url of a finished image back to a vision model with your brand or platform rules. Works for text-to-image and image editing results.
Speech
Summarise and moderate transcripts
Transcribe a call, podcast or stream with WhisperLargeV3Ct2, then ask a model for a summary, chapters or flagged passages. A multi-hour transcript fits in one request on the 1M-token models.
Documents
From scans to structured fields
OCR turns an invoice photo into Markdown. A model with a JSON schema maps it to invoice number, dates, totals and VAT IDs that your code can validate and store.
Agents
Let the model pick the next endpoint
Describe deAPI endpoints as tools. The model answers with a tool call such as text_to_image and its prompt, width and height; your code runs the request and returns the result.
Content
Scripts and prompts for media models
Write the scene list for an image-to-video clip, the voice-over for text-to-speech or product copy from a photo, then send each piece to the right model.
Search
Answers over your own data
Embed documents with Bge_M3_FP16, retrieve the closest chunks and let a model answer from them with sources. Embeddings and chat run on the same key.
Video
Tags and checks from video descriptions
Video analysis describes what happens on screen. A model turns that text into tags, a compliance verdict or a short summary for your catalogue.
Content Moderation with an LLM: Check a Partner Request
A platform that resells generation to partners has to decide, for each request, whether it may run. The call below sends the prompt and the reference image together and gets a verdict your code can act on.
Request
import json
from openai import OpenAI
client = OpenAI(base_url="https://oai.deapi.ai/v1", api_key="YOUR_DEAPI_KEY")
policy = {
"type": "object",
"additionalProperties": False,
"required": ["safe", "categories", "reason"],
"properties": {
"safe": {"type": "boolean"},
"categories": {"type": "array", "items": {"type": "string", "enum": [
"sexual", "minors", "violence", "hate", "self_harm",
"illegal", "weapons", "pii", "none"]}},
"reason": {"type": "string"},
},
}
# reference_image_url: a public HTTPS link or a base64 data URL
check = client.chat.completions.create(
model="qwen3.8-flash",
temperature=0,
messages=[
{"role": "system", "content": "You check requests that partners send "
"to an image generation API. Flag anything that breaks the policy. "
"Return JSON only."},
{"role": "user", "content": [
{"type": "text", "text": "Partner prompt: 'Ad banner: 9mm pistol, "
"no license or background check needed, ships in 24h'"},
{"type": "image_url", "image_url": {"url": reference_image_url}},
]},
],
response_format={"type": "json_schema",
"json_schema": {"name": "moderation", "strict": True,
"schema": policy}},
)
verdict = json.loads(check.choices[0].message.content)
Response: message.content
{
"safe": false,
"categories": ["weapons", "illegal"],
"reason": "The prompt requests an ad banner promoting a firearm
with no license or background check and rapid shipping, which
violates weapons and illegal goods/services policies."
}
usage: 325 prompt tokens, 275 completion tokens
(225 of them reasoning)
Unedited output of qwen3.8-flash from 9 November 2026. The reason is wrapped for width.
The same schema run on glm-5.3 fails, because that model accepts only response_format: json_object. Ask for the keys in the prompt there.
A Moderation Gate in Three Calls
1
Check the input
Prompt and uploaded image go to a vision model with your policy as a JSON schema. Unsafe requests stop here, before any image or video credits are spent.
2
Generate on deAPI
Safe requests go to the image, video or audio endpoint the partner asked for, with the same API key.
3
Check the output
Send the finished image or a few video frames back to the vision model. Deliver, retry or hold for review, and store each verdict next to the request ID.
Chat Models on deAPI and Their Prices
Every model below takes tools, streaming and reasoning_effort. Images go in as a base64 data URL or a public HTTPS link; PDFs, other files and audio are rejected.
Model
Input
Context (tokens)
Max output
Input / 1M tokens
Output / 1M tokens
deepseek-v4.1-flash
Text, image
1,000,000
393,216
$0.165
$0.66
deepseek-v4.1-flash:economy
Text, image
1,048,576
393,216
$0.132
$0.528
glm-5.3
Text
1,000,000
131,072
$1.54
$4.84
glm-5.3-flash
Text, image
1,048,576
131,072
$0.165
$0.55
gpt-6-luna
Text, image
1,050,000
128,000
$0.11
$0.55
hy3
Text
262,144
131,072
$0.1452
$0.5808
kimi-k3
Text, image
1,000,000
131,072
$3.30
$16.50
qwen3.8-flash
Text, image
1,000,000
131,072
$0.165
$0.517
qwen3.8-max
Text, image
1,000,000
131,072
$2.20
$6.60
seed-2.0-lite
Text, image
262,144
131,072
$0.275
$2.20
seed-2.0-mini
Text, image
262,144
131,072
$0.11
$0.44
seed-2.0-pro
Text, image
262,144
131,072
$0.55
$3.30
seed-2.1-turbo
Text, image
262,144
262,144
$0.55
$2.75
Context and output limits from GET https://oai.deapi.ai/v1/models. Prices in USD per million tokens, November 2026. Reasoning tokens are billed as output.
Request and Response
One POST to https://oai.deapi.ai/v1/chat/completions with your deAPI key as the Bearer token. Any OpenAI SDK works once you change the base URL.
Request
model
One of the 13 model IDs above, for example qwen3.8-flash or deepseek-v4.1-flash.
messages
Chat history. A user message can mix text parts with image_url parts on the 11 vision models.
response_format
json_schema with strict: true for a fixed shape, or json_object (the only option on glm-5.3).
tools
Function definitions. The model answers with tool_calls and finish_reason "tool_calls".
stream
true for server-sent events, one chat.completion.chunk per piece.
reasoning_effort
How long the model thinks before it answers, from none to max. glm-5.3 takes only low, high and max.
Response (JSON)
choices[0].message.content
The answer, or the JSON string when you set response_format.
choices[0].message.reasoning_content
The model's reasoning, on models that return it.
choices[0].message.tool_calls
Function name and JSON arguments for your code to run.
usage
Prompt, completion and reasoning tokens for the call.
$5 in free credits
No credit card required
Prepaid, no subscription
Add a Chat Model to Your deAPI Pipeline
Sign up, create an API key and point your OpenAI client at oai.deapi.ai/v1. The same key runs transcription, OCR, image, video and speech.
Yes. Set base_url to https://oai.deapi.ai/v1 and use your deAPI key as the API key. Chat completions, tool calls, JSON output and streaming use the same request and response shapes as the OpenAI SDK.
All except glm-5.3 and hy3, which take text only. Send the image as an image_url part, either as a base64 data URL or as a public HTTPS link that the server can download.
No, those inputs are rejected. Render PDF pages to images or run them through the OCR API first, and transcribe audio with WhisperLargeV3Ct2, then pass the text to the model.
Set response_format to json_schema with strict: true and describe the fields. glm-5.3 accepts only json_object, so list the keys in the prompt when you use it.
Yes, all 13. With tools the model returns tool_calls for your code to run, for example a deAPI image or speech request. With stream: true you get server-sent chunks.
Per input and output token, from the same prepaid balance as the rest of deAPI, with no subscription. Input starts at $0.11 per million tokens; the table above lists every model. New accounts get $5 in free credits, valid for 30 days.