Calculation · list prices of September 21, 2026 to October 3, 2026 · provider prices of October 6, 2026
AI cost calculator: what a month costs.
Pick a workload whose tokens we recorded, or type your own. The calculator prices the same tokens at each model’s list price, with and without prompt caching. It is arithmetic, not a benchmark: it does not say which model would do the work well.
Calculator
Per month · Gemini 3.x Flash
List-price calculation$770
$0.77 per task · 1,000 tasks · SWE-bench-style coding task · with prompt caching
- Uncached input$0.180%
- Cache writes$22229%
- Cache reads$34845%
- Output$20026%
1,000 tasks × (239 in × $0.75 + 295,379 write × $0.75 + 4,639,572 read × $0.075 + 53,353 out × $3.75) ÷ 1M = $770
Across 7 models: $770 (Gemini 3.x Flash) to $9,738 (Claude Fable 5.1) a month. Without prompt caching, Gemini 3.x Flash: $3,901.
The same workload on every model
The tokens stay fixed and only the price changes. Pick a row to price that model above.
Monthly cost at list prices
1,000 tasks a month · SWE-bench-style coding task · with prompt caching
Sorted by monthly cost, cheapest first. Pick a model to price it above.
- Gemini 3.x Flash (Google): $770 per month; selected
- Claude Haiku 4.5 (Anthropic): $1,322 per month; across providers $1,322 to $1,322
- GPT-6.1 Sol (OpenAI): $1,589 per month
- Claude Sonnet 5.5 (Anthropic): $2,643 per month; across providers $2,643 to $2,643
- Claude Opus 5.5 (Anthropic): $4,359 per month; across providers $4,359 to $4,359
- Claude Opus 5 (Anthropic): $6,609 per month; across providers $6,609 to $6,609
- Claude Fable 5.1 (Anthropic): $9,738 per month; across providers $9,738 to $9,738
- Hollow bar: list-price calculation, not a bill
| Model | Per month | Providers: cheapest – priciest |
|---|---|---|
| Gemini 3.x FlashGoogle | $770 | — |
| Claude Haiku 4.5Anthropic | $1,322 | $1,322 – $1,322 |
| GPT-6.1 SolOpenAI | $1,589 | — |
| Claude Sonnet 5.5Anthropic | $2,643 | $2,643 – $2,643 |
| Claude Opus 5.5Anthropic | $4,359 | $4,359 – $4,359 |
| Claude Opus 5Anthropic | $6,609 | $6,609 – $6,609 |
| Claude Fable 5.1Anthropic | $9,738 | $9,738 – $9,738 |
List-price calculation: tokens × price, not a billLinear bars start at $0
7 models priced for the same workload. Lowest Gemini 3.x Flash $770, highest Claude Fable 5.1 $9,738 per month. A list-price calculation, not a bill.
A calculation: the same token mix priced at each vendor’s list price. A different model would use a different number of tokens for the same work.
Assumptions
- A calculation, not a run. Each row prices the same token mix; a different model would use a different number of tokens, and could fail the task.
- Prices are vendor list prices on the dates shown, from the price table in the cost study. Cache writes follow the study’s rule: twice the input price for Anthropic models, the input price for other vendors.
- No batch, volume or subscription discount is applied. Taxes and tool or search fees are not included.
Every model, per task and per month
| Model | $/M in · write · read · out | Per task | Per month | Without caching | Price date |
|---|---|---|---|---|---|
| Gemini 3.x FlashGoogle | $0.75 · $0.75 · $0.075 · $3.75 | $0.77 | $770 | $3,901 | September 21, 2026 |
| Claude Haiku 4.5Anthropic | $1.00 · $2.00 · $0.10 · $5.00 | $1.32 | $1,322 | $5,202 | September 21, 2026 |
| GPT-6.1 SolOpenAI | $2.00 · $2.00 · $0.10 · $10.00 | $1.59 | $1,589 | $10,404 | October 3, 2026 |
| Claude Sonnet 5.5Anthropic | $2.00 · $4.00 · $0.20 · $10.00 | $2.64 | $2,643 | $10,404 | September 21, 2026 |
| Claude Opus 5.5Anthropic | $4.00 · $8.00 · $0.20 · $20.00 | $4.36 | $4,359 | $20,808 | September 21, 2026 |
| Claude Opus 5Anthropic | $5.00 · $10.00 · $0.50 · $25.00 | $6.61 | $6,609 | $26,010 | September 21, 2026 |
| Claude Fable 5.1Anthropic | $10.00 · $20.00 · $0.25 · $50.00 | $9.74 | $9,738 | $52,020 | September 21, 2026 |
| Jev 1.13 (router)TypeSafe · router only: it cannot do this work | $0.042 · $0.042 · $0.0042 · $0.00 | $0.032 | $31.90 | $207 | September 23, 2026 |
How the prices and token mixes were recorded Calculate the cache break-even turn
Calculation · fixed recorded inputs
Agency team presets
Example assumptions: 10 tasks to finish per developer per workday, 21 workdays, and one routing decision per model call. These are assumed team volumes.
Tasks to finish: 1,050. Expected attempts: 1,386. Expected routing decisions: 68,544.
No added router
- Monthly total
- $3,890.78
- Per developer
- $778.16
Claude Sonnet 5.5 router
- Monthly total
- $4,233.22
- Per developer
- $846.64
Jev 1.13 router
- Monthly total
- $3,893.09
- Per developer
- $778.62
Agency cost calculation details
| Routing add-on | Agent cost | Added routing cost | Monthly total | Per developer |
|---|---|---|---|---|
| No added router | $3,890.78 | $0.00 | $3,890.78 | $778.16 |
| Claude Sonnet 5.5 router | $3,890.78 | $342.45 | $4,233.22 | $846.64 |
| Jev 1.13 router | $3,890.78 | $2.31 | $3,893.09 | $778.62 |
Recorded inputs: 25/33 tasks resolved, $92.63751 total notional cost and 1,632 model calls. Prices used in the example: Agent 2026-09-21; Sonnet router 2026-09-21; Jev 2026-09-23. Sonnet routing costs $0.004996 per decision; Jev costs $0.0000337.
Expected independent retries hold the mean cost and pass rate fixed. We did not run these teams or retries. Router choices could change the model, tokens and outcomes; this calculation holds them fixed. Subscription prices, developer time and review time are unknown here. No cost estimate is an invoice.
Read the recorded inputs and formula. The token calculator above uses attempts and a fixed token mix; these presets count tasks to finish.