Personal
What will this cost me.
Pick a task, a volume and a preference. Prices are models.dev list prices. The winner is labelled by the preference you set, not by a hidden score. The table is 1,000 coding edits at the list price.
| Model | Input / 1M | Output / 1M | 1,000 coding edits |
|---|---|---|---|
| DeepSeek V4 Flash | $0.15 | $0.60 | $1.80 |
| Gemini 4 Argon | $2.00 | $10.00 | $27.00 |
| Muse Wonder | $4.00 | $20.00 | $54.00 |
| Mistral Large 4 | $1.36 | $4.18 | $14.43 |
| GPT-6.1 Sol | $2.00 | $10.00 | $27.00 |
| Claude Sonnet 5.5 | $2.00 | $10.00 | $27.00 |
| Claude Opus 5.5 | $4.00 | $20.00 | $54.00 |
| GPT-6 Luna | $0.10 | $0.50 | $1.35 |
“Coding edit”: A file snippet in, a patch out. The token counts are a planning assumption you can edit. They are not a measured trace. List price is one rate; this snapshot has no tier schedule.
Cache off: every input token uses the input list price. Batch is off. Turning it on does not invent a discount.
About 0 tokens if you count 4 characters as one token. This is not the model’s tokenizer.
Stack bill
Add consumer plans whose fetched HTML states a monthly price. Annual and promo figures are left out. This is separate from token cost. Seats are not multiplied.
Apps $0.00 + Gemini 4 Argon $27.00 = $27.00
Team API bill
Same list price as the top row above, times 10 people each running the task 1000 times a month. Cache follows the toggle. It is not an invoice.
Gemini 4 Argon: $270.00 / month · $3,240.00 / year. Trust notes