perceptix-vex-amber-2.0
Arioron model.
- Context
- 1,048,576
- Max output
- 65,536
- Input
- text · image
- Reasoning
- Yes
Arioron · VexAPI
One API for Arioron models: reasoning, vision, documents, tool calling, structured output, embeddings and agents. It speaks the OpenAI and Anthropic formats, so the SDK you already use works by changing one line.
from openai import OpenAI
client = OpenAI(base_url="https://api.arioron.com/v1", api_key=VEX_API_KEY)
response = client.responses.create(
model="vex-photon-lite",
input="Explain quantum entanglement to a curious "
"teenager in two sentences.",
)
print(response.output_text)
import OpenAI from "openai";
const client = new OpenAI({ baseURL: "https://api.arioron.com/v1", apiKey: VEX_API_KEY });
const response = await client.responses.create({
model: "vex-photon-lite",
input: "Explain quantum entanglement to a curious " +
"teenager in two sentences.",
});
console.log(response.output_text);
curl https://api.arioron.com/v1/responses \
-H "Authorization: Bearer $VEX_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "vex-photon-lite",
"input": "Explain quantum entanglement to a curious teenager in two sentences."
}'
Output · captured from a real request
Imagine you have two special coins that are linked, so that if one lands on heads, the other instantly lands on tails, no matter how far apart they are. Quantum entanglement is like that, but with tiny particles whose fates become intertwined, so that measuring one tells you something definite about the other, even if they're light-years away.
Scroll
Compatibility
VexAPI implements the OpenAI Chat Completions and Responses formats and the Anthropic Messages format, verified against both official SDKs: streaming, tool calling, structured output, vision and error handling.
Point the client at VexAPI and choose an Arioron model. Nothing else changes.
- client = OpenAI(base_url="https://api.openai.com/v1")+ client = OpenAI(base_url="https://api.arioron.com/v1") client.chat.completions.create( - model="gpt-…",+ model="perceptix-vex-amber", messages=messages, tools=tools, stream=True)
- client = Anthropic(base_url="https://api.anthropic.com")+ client = Anthropic(base_url="https://api.arioron.com") client.messages.create( - model="claude-…",+ model="perceptix-vex-amber", max_tokens=1024, messages=messages, tools=tools)
Platform
Everything between a first request and a production AI product, on one key, one bill and one set of logs.
Typed outputs, server-side conversation state and built-in web search.
/v1/responses
02
Describe functions in JSON Schema; the model returns calls with validated arguments.
tools
03
Output guaranteed to parse and match your schema. No retries, no regex.
json_schema
04
Read images, PDFs and text from URLs, data URLs or uploaded files.
vision
05
Multi-step runs with permissioned tools, timeouts, cancellation and a full trace.
/v1/agents
06
Normalized vectors with adjustable dimensions for search and retrieval.
/v1/embeddings
07
Thousands of requests from one JSONL file, processed durably in the background.
/v1/batches
08
Signed notifications when agent runs and batches finish, with retries.
/v1/webhooks
09
Score models against your own test cases before you ship a change.
/v1/evals
Models
Long context, multimodal input and tool use across the lineup. Choose depth or speed per
request with reasoning_effort.
Arioron model.
Strongest Amber Model Yet
Latest series of Photon Lite models
Text embeddings for search, clustering and retrieval.
Observability
Every request is logged with its model, latency, status and exact token counts, measured by the
model service, never estimated. Every response carries an x-request-id you can
paste into your logs.
Production
The parts you only notice when they're missing: scoped keys, predictable limits, safe URL handling and background work that survives a restart.
Keys
Project-scoped, read-only and expiring keys, accepted in headers only, never in URLs.
Limits
Per-key rate limits, reported in headers on every response, with a Retry-After when you hit one.
Network
Images and pages are fetched from public addresses only. Internal networks are refused.
Jobs
Batches and agent runs survive restarts and retry automatically. Webhooks are HMAC-signed.