Nemotron 3 Ultra 550B API

Pricing

₹80/1M input · ₹330/1M output. Pay-as-you-go micro-metered billing. UPI and cards accepted with zero minimum commitment.

Good fit for agentic coding

• Autonomous coding agents that read files, run tools, and iterate
• Multi-step workflow automation where the model calls your functions
• Long repository or document context that must stay resident across turns
• Internal developer tools that need reliable structured output

OpenAI-Compatible Integration

Drop into existing OpenAI SDK code by changing the base URL to https://api.luminoai.co.in/v1. The request shape below is the agentic coding pattern.

curl https://api.luminoai.co.in/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $LUMINO_API_KEY" \
  -d '{
    "model": "nemotron-3-ultra-550b",
    "messages": [
      {"role": "system", "content": "You are a helpful assistant."},
      {"role": "user", "content": "Read the failing test, locate the bug, and return a patch."}
    ]
  }'

Model Specifications

Why agentic workloads get expensive

Agent loops resend the whole conversation on every turn, and a failed tool call bills the entire history again. Three controls matter most: cap the loop iterations, trim tool output before it re-enters history, and set a realistic max_tokens instead of leaving the ceiling open.

Nemotron 3 Ultra 550B API FAQ

Does Nemotron 3 Ultra 550B support function calling?

Yes — send tools in the standard OpenAI tools array and the model returns tool calls in the same shape your existing client already parses.

What dominates the cost of an agent loop?

Input tokens, because the full history is resent every turn. Cached input pricing is what keeps long loops affordable.

How do I keep long sessions from drifting?

Keep the system prompt stable so the prefix stays cacheable, and summarise older turns instead of dropping them silently.