One prompt.
The best model for the job.
Aridel classifies every prompt and sends it to the right frontier model — Claude, GPT, Gemini, Kimi, DeepSeek, and more. Pay pennies for the easy stuff. Reach for the smartest model only when you actually need it.
Smart routing
An embedding-based classifier reads each prompt and picks the model with the best price/quality fit. Easy questions go to cheap, fast models. Hard ones get the heavyweights.
56 models, one thread
The GPT-5 family, Claude (Opus, Sonnet, Haiku), Gemini 3, plus the strongest open-weight models — DeepSeek, Kimi, Qwen, MiniMax, GLM, Amazon Nova — behind one chat UI and one OpenAI-compatible API.
Stop overpaying
Most prompts don't need a top-tier model. Routing trims the bill on the long tail of easy queries — without you having to think about which model to pick.
How it works
Three steps, no model picking, no per-provider accounts.
You write a prompt
Same flow as ChatGPT or Claude — one box, one stream. Attach files, toggle web search, keep your conversation history.
Aridel scores every model
The router embeds the prompt, weighs it against 45 benchmark-backed question categories, and scores all 56 models on expected quality versus price. Three auto modes — eco, balanced, max — set how much a quality point is worth to you.
Streamed answer, every time
The chosen model streams back token-by-token. We log which model was picked and why, so you can audit the routing decision later in Insights.
Also an API you can build on
The web app runs on the exact same public API you get. It's
OpenAI-compatible — point your existing client at
api.aridel.ai, set model="auto", and
routing is on. Or pip install aridel:
from aridel import Aridel
client = Aridel(api_key="aridel_live_...")
response = client.chat.completions.create(
model="auto", # let the router pick
messages=[{"role": "user", "content": "Hello!"}],
)
print(response.choices[0].message.content)
print(response.routing) # which model, and whyRouting as a service
POST /v1/route returns the routing decision —
chosen model, complexity, confidence, cost estimate — without
calling any model. Batch hundreds of prompts in one call and
drive your own stack with it.
Real workloads
Tool calls, vision, file inputs, pre-flight token counting, per-request model pools, and async batch jobs on supported providers' discounted batch endpoints.
Insights & audit logs
Every routed call feeds your Insights dashboard — model mix,
savings estimates, quality lift. Opt-in audit logging keeps a
write-once record per request, readable via
/v1/audit-logs.
Pricing
A plan sets your monthly routing volume. How the models get paid is up to you: bring your own provider keys, or use Aridel credits.
Free
$0 / month
up to 25M tokens / month
- Auto-routing, all three modes
- Basic insights
- Starter credits — no card required
Team
$29 / month
up to 300M tokens / month
- Everything in Free
- Full insights depth
- Write-once audit logs
- Email support
Scale
$149 / month
up to 3,000M tokens / month
- Everything in Team
- Priority support
- Higher rate limits
Bring your own keys — $0 per token
Use your own provider API keys: you pay each provider directly and Aridel adds nothing on tokens, ever. Pin any model in the catalog or keep auto-routing. Keys are encrypted with AWS KMS, or passed per-request and never stored — full control over provider, region, and data jurisdiction.
Aridel credits — tokens at cost + 15%
No provider accounts needed: a prepaid wallet on Aridel's own keys. You pay exactly what the providers charge, plus a flat 15% — cache discounts included, nothing hidden. Top up $1–$100 at a time; credits never expire.
Plan volume counts all tokens a request bills — input, cached context, and output. You get a warning at 80% of your monthly cap and a hard stop at 100%; upgrading lifts it immediately. Open-weight models may be served by their native platforms — Z.AI, Alibaba Cloud, or Moonshot AI (China) — at the best available price; bring your own keys to control exactly which providers serve you.
Choose your view
One site, three finishes. Your pick is remembered on this device and applies to every page.