Cost Optimization

Goldnat provides several levers to control costs. This guide explains strategies for getting the most value from your credits.

Choose the Right Model for the Task

Not every task needs the most powerful model. Match model capability to task complexity:

Task	Recommended Model	Why
Simple data retrieval	Claude Haiku 4.5	Lowest cost, fast, sufficient for reading and formatting data
Content summarization	Claude Haiku 4.5	Good comprehension at very low cost
General analysis	Claude Sonnet 4.6	Balanced quality and cost
Complex reasoning	Claude Opus 5	Highest reasoning capability, 1M token context
Code generation	Claude Sonnet 4.6 / Opus 5	Strong code quality at different price points

Model Cost Comparison

All models on Goldnat are from Anthropic. Costs are shown in USD per million tokens — the exact credit cost for each interaction is displayed in the chat UI.

Model	Input ($/MTok)	Output ($/MTok)	Context	Tier
Claude Haiku 4.5	$0.80	$4	200K	Very Low
Claude Sonnet 4.6	$3	$15	200K	Medium
Claude Opus 4.7	$3	$25	200K	High
Claude Opus 5	$3	$25	1M	High

A task using Claude Opus 5 for output costs over 6x more than Claude Haiku 4.5. Choosing the right model can dramatically reduce spending.

TIP

The platform may have additional model variants available (e.g., dated snapshots). Check the model selector in Chat for the full list and live credit costs.

Use BYOK to Avoid Credit Markup

Platform credits include a markup over raw provider costs. With BYOK, you pay the provider directly at their wholesale rates:

• Without BYOK: 1,000 output tokens with Claude Sonnet 4.6 = platform credits (including markup)
• With BYOK: 1,000 output tokens with Claude Sonnet 4.6 = $0.015 (direct Anthropic rate, no markup)

If you regularly spend $50+/month in credits, BYOK can save 50-80% depending on the model.

See the BYOK Guide for setup.

Auto Top-Up vs. Manual Purchase

Approach	Pros	Cons
Manual purchase	Full control, buy only when needed	Risk of running out mid-task
Auto top-up	No interruptions, always have credits	May overspend if not monitored

Recommended Auto Top-Up Strategy

Set conservative limits:

• Threshold: 100 credits
• Amount: 500 credits
• Monthly limit: 2,000 credits

This prevents runaway spending while ensuring you never hit zero during a conversation.

Monitor Usage in Analytics

The Analytics page shows:

• Total credits consumed per day
• Credits broken down by model
• Token usage trends

Review analytics weekly to identify:

• Unexpected spikes — a misconfigured agent running too frequently
• Model waste — using an expensive model for tasks a cheaper one could handle
• Credit trends — whether your usage is growing and you need a plan upgrade

Set Agent Round Limits

Each agent round costs credits. A runaway agent can consume your entire balance.

Cost estimation per round (approximate):

Input tokens: ~1,500 (system prompt + context + tool results)
Output tokens: ~500 (AI response + tool calls)

Claude Haiku 4.5:  ~$0.003 per round (very low)
Claude Sonnet 4.6: ~$0.012 per round (moderate)
Claude Opus 5:     ~$0.017 per round (higher)

An agent with maxRounds: 10 using Claude Opus 5 running hourly can cost over 5x more than the same agent using Claude Haiku 4.5. Multiply by 24 hours and 30 days — the difference adds up fast.

Always start with low maxRounds and increase only if needed.

Keep Conversations Focused

Credit costs scale with conversation length because the entire history is sent as input tokens on each message. Strategies:

1. Start new conversations for new topics instead of continuing old ones
2. Be concise in your messages
3. Clear history periodically if you do not need old conversations

The platform loads the last 20 messages as context. Longer conversations still cost more per message because each message includes the prior 20 messages as input.

Optimize Agent Prompts

A shorter, focused system prompt reduces input tokens per round:

Before (230 tokens):

You are a highly intelligent and capable AI assistant that has 
been designed to help users manage their online forums. You 
should always be polite, thorough, and comprehensive in your 
responses. Please search for new posts and provide a summary.

After (45 tokens):

Search for forum posts from the last hour. Summarize: title, 
author, category. Output as a Markdown table.

The "after" version costs ~80% less in input tokens every round while being more precise.

Summary of Strategies

Strategy	Potential Savings	Effort
Use cheaper models	10-12x	Low — change model selection
BYOK	50-80%	Low — one-time setup
Lower agent rounds	50-90%	Low — adjust one setting
Focused conversations	20-40%	Medium — change workflow habits
Optimized prompts	10-30%	Medium — rewrite prompts
Monitor analytics	Prevents waste	Low — weekly check