What a month of coding requests looks like
Coding tools send a lot of input: the files you have open, the chat so far, and the tool's own instructions. The output is a few hundred lines at most. The example is 6,600 requests a month (300 a working day), about 15,000 tokens in and 1,000 out each.
The same month on each model
At list price with no caching, grouped by provider, largest model first.
| Model | Input | Output | Per month |
|---|---|---|---|
| Claude Opus 5.5 | $396 | $132 | $528 |
| Claude Sonnet 5.5 | $198 | $66.00 | $264 |
| Claude Haiku 5.5 | $9.90 | $3.30 | $13.20 |
| GPT-5.5 | $495 | $198 | $693 |
| GPT-6.1 Sol | $198 | $66.00 | $264 |
| GPT-6.1 Terra | $99.00 | $33.00 | $132 |
| GPT-5 mini | $24.75 | $13.20 | $37.95 |
| Gemini 3.1 Pro Preview | $198 | $79.20 | $277 |
| Gemini 3.8 Flash | $74.25 | $24.75 | $99.00 |
| Grok 4.7 | $198 | $39.60 | $238 |
| Grok Build 0.1 | $99.00 | $13.20 | $112 |
The lowest list price here is Claude Haiku 5.5 at $13.20. Gemini rows use the launch price and are for comparison; FUSR.ai does not route to Gemini.
Caching changes the order
Because coding tools resend the same files every turn, much of the input can come from the cache. With 70 in 100 input tokens cached, Sonnet 5.5 goes from $264 to $132 a month. Cache discounts differ by provider, so the ranking moves.
| Model | List price | 70 in 100 input tokens cached | Batch price |
|---|---|---|---|
| Claude Opus 5.5 | $528 | $265 | $264 |
| Claude Sonnet 5.5 | $264 | $132 | $132 |
| Claude Haiku 5.5 | $13.20 | $6.96 | $6.60 |
| GPT-5.5 | $693 | $381 | $347 |
| GPT-6.1 Sol | $264 | $132 | $132 |
| GPT-6.1 Terra | $132 | $66.16 | $66.00 |
| GPT-5 mini | $37.95 | $22.36 | $18.98 |
| Grok 4.7 | $238 | $134 | Not listed |
| Grok Build 0.1 | $112 | $56.76 | Not listed |
Anthropic cache writes are left out, so the Claude cached rows are a floor.
When a bigger model is worth it
A cheap model that needs three tries costs more than a bigger one that gets it right once, and your time costs more than either. A pattern that works for many people: a small model for quick questions, renames, and tests, and a large one for multi file changes and hard bugs.
You can wire that up yourself, or point your editor at an AI API router that picks per request. Setting it up in Cursor is in how to use your own API key in Cursor.