Why reasoning models cost more
How thinking tokens raise the cost of reasoning models, with live prices for low and high thinking variants.
Estimated monthly cost — Reasoning request
Calculated live from current TokenRex prices.
| Model | Input / 1M | Output / 1M | Per month |
|---|---|---|---|
| GPT-6 Sol Low | $2.00 | $10.00 | $170.00 |
| GPT-6 Sol | $2.00 | $10.00 | $170.00 |
| GPT-6 Sol Deep | $2.00 | $10.00 | $170.00 |
| GPT-6 Sol Max | $2.00 | $10.00 | $170.00 |
The workload we priced
5,000 requests a month, each with about 2,000 input tokens and 3,000 output tokens. One token is roughly three quarters of an English word.
What drives the bill
Thinking is billed as output tokens, so higher thinking levels raise output usage.
Use low thinking for routine tasks and high thinking only where accuracy matters.
Questions
- Are these exact monthly bills?
- They are estimates from live TokenRex prices and the token counts shown. Real bills depend on your actual prompt and reply lengths, which every API response reports in its usage field.
- How do I lower the cost?
- Shorten system prompts, cap reply length, cache repeated input where the model supports it, and send simple requests to a smaller model.
- Which currency am I billed in?
- Wallet top-ups and billing are in Indian rupees. Prices display in US dollars by default, and signed-in users can choose another display currency.
Related
- What is a token?
- What is prompt caching?
- How to reduce LLM API costs
- How much does an AI chatbot cost per month?
- How much does an AI coding agent cost?
- How much does a RAG app cost to run?
- Cost to summarise documents with an LLM
- Cost of AI translation with an LLM API
- Cost to classify emails and tickets with AI
- Cost of extracting structured data with an LLM