What one run costs
A typical run of this task sends about 8K tokens and gets back 1.2K, over 4 calls — 36.8K tokens in total. Those are the numbers below, against the published prices.
| Model | Vendor | Per run | Can do what this needs? | Price evidence |
|---|---|---|---|---|
| GPT-5 Nano | OpenAI | $0.00352 | not stated | Vendor page |
| Gemini 2.5 Flash Lite | $0.00512 | not stated | Vendor page | |
| Claude Haiku 5.5 | Anthropic | $0.00560 | confirmed | Vendor page |
| GPT-6 Luna | OpenAI | $0.00560 | not stated | Vendor page |
| GPT-5.6 Luna | OpenAI | $0.0122 | not stated | Vendor page |
| GPT-5.4 Nano | OpenAI | $0.0124 | not stated | Vendor page |
| Gemini 3.1 Flash Lite | $0.0152 | not stated | Vendor page | |
| GPT-5 Mini | OpenAI | $0.0176 | not stated | Vendor page |
Paste the actual rows, not a description of them. A model guessing at your column names will write a formula that looks right and references a column that does not exist.
Formulas are a reasoning task, and this is one of the places where the cheapest tier genuinely underperforms: an off-by-one in a range or a misplaced absolute reference is the classic failure, and it is silent.
Ask it to explain what the formula does in words before you paste it in. That single step catches most of the wrong ones.
What we would pick
This section is our judgement, not a figure read off a page. Everything above is arithmetic on published prices; this is an opinion, and it is labelled as one.