Two models can have the same input price and still cost very different amounts a month. Comparing them well comes down to three questions: what a month of your work costs, how fast each one answers, and how hard it is to switch later.
Key takeaways
- Compare on a month of real work, not on one price per token.
- For chat and agents, the output price usually matters most.
- Speed is a range: look at the typical answer and the slow one.
- On one gateway, switching models is one word in your code.
A worked example: Claude Sonnet 5.5 vs GPT-6.1 Terra
Both are current, strong all-round models. Here they are side by side, at your price on withConflux (the provider's price plus a small margin), per 1M tokens:
| Claude Sonnet 5.5 | GPT-6.1 Terra | |
|---|---|---|
| Input, per 1M tokens | $2.50 | $2.50 |
| Output, per 1M tokens | $12.50 | $9.38 |
| Context | 1M tokens | 1.05M tokens |
| Can use | vision, tools, reasoning | vision, tools, JSON, reasoning |
The input price is the same. The output price is not, and output is where chat and agent bills grow.
1. What does a month cost?
Take a support chatbot with 30,000 messages a month, about 1,500 tokens in and 300 out each. That is 45M tokens in and 9M out:
| Model | Input | Output | A month |
|---|---|---|---|
| Claude Sonnet 5.5 | $112.50 | $112.50 | $225.00 |
| GPT-6.1 Terra | $112.50 | $84.42 | $196.92 |
Change the volume and the token counts to your own on the Sonnet 5.5 vs Terra page: it works out three workloads and has a calculator.
2. How fast does it answer?
A vendor chart tells you how a model did in their test. What matters is how it answers your calls. We measure answer time on our gateway from live traffic and show it as a range: the typical call, the slow one, and how soon the first words arrive. See API speed for how it is measured.
3. Can you switch later?
Through withConflux, OpenAI and Anthropic models sit behind one OpenAI-compatible endpoint. Switching is the model name:
client.chat.completions.create(model="gpt-6.1-terra", messages=messages)
# or model="claude-sonnet-5-5"Try both on the same prompts before you decide. With one key you can run the same test against each model and compare the answers, the cost per call and the time.
Pick by the work, not by the name
Pick Claude Sonnet 5.5 if your prompts and tools are already tuned for Claude, or you want Anthropic next to OpenAI. Pick GPT-6.1 Terra if output is a big part of your bill or you need JSON mode. Every other current pair is on the compare page, with the date each number was checked.
Not sure what your own usage would cost? Upload your bill for a free check, or compare any two models.