LumeAPISupport agent
50M input + 15M output tokens / month
5–8 hops per resolved ticket
- Official (Claude Sonnet 4.6)
- $375.00/mo
- LumeAPI
- $225.00/mo
- Monthly savings
- $150.0040% off
AI Agent API
AI Agent API — Multi-Model, Cost-Controlled Agent Loops
Build agents on GPT, Claude, and Gemini with lower per-hop token costs — switch models per step without three integrations.
Agents multiply tokens per user request. Lower catalog rates directly reduce cost-per-completed-task.
- OpenAI-compatible
- Pay as you go
- One API key
- GPT, Claude & Gemini
Agent-friendly model pricing
Official reference vs LumeAPI catalog rates. Pricing unit: per 1M input / output tokens. Last updated: July 2026. Source: provider list price.
Monthly cost examples
Illustrative totals for claude-sonnet-4-6 using catalog list prices — your actual bill depends on retries, tool loops, and output length.
Coding agent
120M input + 40M output tokens / month
Tool loops with long context
- Official (Claude Sonnet 4.6)
- $960.00/mo
- LumeAPI
- $576.00/mo
- Monthly savings
- $384.0040% off
Agents multiply cost per user action
A single user request in an agent product might trigger five to fifteen model calls with tools. A 40% lower per-hop rate changes unit economics more than shaving latency milliseconds.
Route cheap classification and planning to Flash or mini models; reserve Sonnet, Sol, or Opus for steps where errors cascade. Log model id per hop in your orchestrator and reconcile with LumeAPI Usage.
Read the linked Research article for formulas and scenarios; this commercial page is the conversion path to create a key and run first agent traffic.
Self-serve path: register to first API call
LumeAPI is designed for developers who want to integrate without scheduling demos. Create an account, confirm your email, and open Console to generate an API key. Fund your USD wallet with USDT on supported chains when you are ready for billable traffic—there is no mandatory minimum beyond what your tests require.
Point your OpenAI-compatible client at https://api.lumeapi.site/v1, set Authorization to Bearer your key, and pass a catalog model id in the model field. Run a short curl or SDK script from /docs to verify latency, streaming, and error handling before you attach the key to production services.
Use Usage logs to reconcile per-call cost with finance forecasts. When a model tier is too expensive or quality is insufficient, change model id—not your entire integration. For cross-provider price tables and Research deep dives, follow internal links on this page rather than duplicating migration math here.
Documentation, catalog, and support
Every catalog model has a detail page under /models with official reference pricing, LumeAPI pricing, and links to /docs/models/{id} for parameters and curl examples. Start there when this commercial page points you to a model id you have not called before.
The /docs index lists gateway authentication, Chat Completions, image endpoints, and async video patterns. llms.txt bundles the same information for agent tooling—useful when you want a single URL to paste into Cursor or an internal bot.
Research articles explain why bills grow and how to compare providers; commercial pages like this one explain what LumeAPI offers and how to start. Follow internal links instead of searching for duplicate migration content across pages.
If billing, chain deposits, or integration behavior is unclear, use /contact for support channels. Include your model id, approximate request time, and whether the issue is authentication, balance, or model parameters—that speeds up resolution.
Why LumeAPI
Mainstream models only
GPT, Claude, and Gemini catalog ids with published per-token or per-media rates — not obscure small models marketed as discounts.
Transparent pricing
Official reference vs LumeAPI columns on every commercial page. Your request model id matches Usage logs and billing.
OpenAI-compatible gateway
One base URL for Chat Completions, image generations, and async video. Swap key, base URL, and model id — keep your SDK.
Self-serve wallet
Top up with USDT, create keys in Console, and track per-call cost in Usage — no sales calls required.
Get started in three steps
- Create an API key — register and open Console.
- Set the LumeAPI base URL — https://api.lumeapi.site/v1
- Choose a supported model id — from the table above or model catalog.
Migrate in minutes
Three values change: API key, base URL, model id. Everything else stays the same.
Before (expensive per-hop agent)
from openai import OpenAI
client = OpenAI(api_key="YOUR_OPENAI_API_KEY")
response = client.chat.completions.create(
model="claude-sonnet-4-6",
messages=[{"role": "user", "content": "Hello"}],
)After (LumeAPI)
from openai import OpenAI
client = OpenAI(
api_key="YOUR_LUMEAPI_KEY",
base_url="https://api.lumeapi.site/v1",
)
response = client.chat.completions.create(
model="claude-sonnet-4-6",
messages=[{"role": "user", "content": "Hello"}],
)Full step-by-step rollout, streaming checks, and FAQ: AI Agent API Bills Out of Control? →
Migration & compatibility
What changes
API key, base URL (https://api.lumeapi.site/v1), and model id to a LumeAPI catalog entry. Message shape stays OpenAI-compatible for most apps.
What to test
Streaming, tool calling, JSON mode, and error handling on your heaviest models. Shadow 5–10% of traffic before full cutover.
What may differ
Provider-native features (Anthropic Batch, Google Grounding, OpenAI Assistants) may require the official API. Test your exact payload.
Rollback
Keep environment variables for base URL and model id. Switch back instantly if staging tests fail.
Trust & billing
Tool calling?
Supported on catalog text models — test your schema.
Model per hop?
Change model id each request in your orchestration code.
Billing transparency?
Usage logs show model id and cost per hop.
Silent model swap?
No. Exact model id in request and logs.
Built for these workloads
Model routing
Flash for cheap steps, Sonnet or Sol for hard reasoning.
Tool calling
Test tools on catalog models via Chat Completions.
High consumption
40–50% lower rates compound across every agent hop.
Cost per task
Measure completed workflows, not single API calls.
Related guides
FAQ
Best API for agents?
Depends on workload. Compare Sonnet, Sol, and Flash on cost per completed task.
Multi-model for agents?
Yes — one key, swap model id per agent step.
Agent bills out of control?
See linked Research article for formulas and routing tips.
Streaming in agents?
Yes — test stream resume in staging.
High volume?
See /high-volume-llm-api.
Coding agents?
See /coding-ai-api.
Cut your agent API bill
Route agent loops through https://api.lumeapi.site/v1. Need help choosing a model? Browse the developer docs or contact support.