GPT-6.2 Astra ProOpenAI
- Input not in cache
- $720
- Output
- $720
- Total per month
- $1,440
- Per day
- $48.00
- Per request
- $0.048
Describe your traffic once, or start from a quick example, and read the monthly bill on up to four models line by line. Then see what switching saves, where the cheaper model changes, and how long your budget lasts as traffic grows.
Lowest of your selection: Google Gemini 4 Neon, at $540 a month.
Every figure comes from the price each source publishes for the model: nothing is assumed where a price is missing. A 30-day month is used throughout.
Collected from the OpenRouter models API (476 entries). 104 models are cross-checked against provider list prices in the LiteLLM price table.
Every model in your selection against one baseline: price per token, monthly cost and the difference in money and percent. Then the prompt and answer lengths at which the cheaper model flips.
Switching from GPT-6.2 Astra Pro to Gemini 4 Neon saves $900 a month (−63%), or $10,800 over 12 months at this volume.
| Model | Input, per 1M tokens | Output, per 1M tokens | Effective inputwith 0% from cache | Cost per month | Against GPT-6.2 Astra Pro |
|---|---|---|---|---|---|
| GPT-6.2 Astra Probaseline | $8.00 | $40.00 | $8.00 | $1,440 | baseline |
| Claude Fable 5.2Anthropic | $8.000% | $40.000% | $8.00 | $1,440 | same cost |
| Gemini 4 NeonGoogle | $3.00−63% | $15.00−63% | $3.00 | $540 | saves $900−63% |
Percentages under each price compare it with the baseline’s price. Effective input is the average price of one million input tokens once your cached share is billed at the model’s cache-read price (or at the normal input price when none is published).
GPT-6.2 Astra Pro vs Claude Fable 5.2
Same effective prices: they cost the same at any length.
GPT-6.2 Astra Pro vs Gemini 4 Neon
Gemini 4 Neon is cheaper at every prompt and answer length: its effective input price and its output price are both lower or equal.
Each point moves one length and keeps the other, your cached share and your volume as they are. Volume never changes the winner: it scales both bills equally.
| Output tokens per request | GPT-6.2 Astra Pro | Claude Fable 5.2 | Gemini 4 Neon |
|---|---|---|---|
| 100 | $840 | $840 | $315 |
| 300 | $1,080 | $1,080 | $405 |
| 600(yours) | $1,440 | $1,440 | $540 |
| 1,000 | $1,920 | $1,920 | $720 |
| 3,000 | $4,320 | $4,320 | $1,620 |
| 10,000 | $12,720 | $12,720 | $4,770 |
Lowest cost in each row in bold. Requests, input tokens and cached share stay as set above.
Set the monthly budget for this workload and how fast your traffic grows. For each model you see the share of the budget it uses, what is left, how many requests a day the budget pays for and the month the bill would pass it.
1 of the 3 priced models in your selection fit a budget of $1,000 a month at today’s volume of 1,000 requests a day.
Month 1 is today’s bill. Growth compounds on the number of requests; prompt and answer lengths and the cached share stay as set above, and prices stay at the latest collection. The 12-month total adds twelve 30-day bills.
An API bill for a language model has three lines. Input is everything you send: instructions, conversation history, retrieved documents. Output is what the model writes, at a price that is usually several times higher. Cached input is the part of your input the provider has already seen and can re-read at a discount.
Monthly requests are requests per day times 30. Input tokens per month are monthly requests times input tokens per request; the cached share of them is billed at the cache-read price, and the rest at the input price. Output tokens per month are monthly requests times output tokens per request, billed at the output price. Each product is divided by one million, because prices are quoted per million tokens.
If you already call an API, read the token counts from the usage field of a few real responses and average them: guesses are usually off by a wide margin. If you are planning, start from one of the scenarios and adjust. For chat, remember that every turn re-sends the conversation so far, so input grows with the length of the dialogue.
The total is an estimate of the usage charge only. It leaves out cache-write fees, tool and search fees, images and audio, taxes and any committed-use discount. Use it to compare models on equal terms and to see which of the three lines dominates your bill, then confirm on the provider’s pricing page.
The savings are the difference between two monthly bills, divided by the baseline’s bill for the percentage; the 12-month figure is that difference times twelve. A break-even point is where two models’ costs per request are equal. Holding the prompt fixed, the cost of each request is a straight line in the answer length, so two models swap places at most once: at the input-cost gap divided by the output-price gap. The same works the other way, holding the answer fixed and moving the prompt. Volume multiplies both bills by the same factor, so it never changes the winner.
Cost per request is the monthly bill divided by the requests in a 30-day month. The requests a day your budget pays for are the budget divided by 30 times that cost, rounded down. With monthly growth, the bill of month n is today’s bill times (1 + growth) to the power n − 1, and the planner reports the first month it passes the budget, looking up to 36 months ahead.
Requests per day times 30 gives the requests in a month. Those are multiplied by the input and output tokens per request, divided by one million and multiplied by the model’s price per million tokens. The share of input you mark as cached is billed at the model’s cache-read price when the source lists one.
For English text, a token is about four characters, so 1,000 tokens are roughly 750 words and a full page of prose is 600 to 700 tokens. Code, other languages and long numbers use more tokens per word. Tokenisers differ between providers, so treat any estimate as a range.
Cache-write charges, per-request fees, reasoning tokens that some models bill as output, image and audio inputs, web-search tool fees, taxes and volume discounts. For reasoning models, raise the output tokens to include the hidden reasoning your requests usually produce.
Because the source lists no cache-read price for them. The calculator then bills cached input at the normal input price instead of assuming a discount the provider may not offer.
Yes. The numbers you enter, the models you pick, the baseline, the budget and the growth rate are written into the address of the page. Copy the link and whoever opens it sees the same calculation with the latest collected prices.
Each example only fills in the four numbers, which you can then change. Support chat: 1,000 requests a day, 3,000 input and 600 output tokens each. Document summaries: 17 requests a day, 6,700 input and 400 output tokens each. Coding assistant: 1,000 requests a day, 6,000 input and 800 output tokens each, 70% of the input from cache. Q&A over documents: 333 requests a day, 4,000 input and 350 output tokens each. Chatbot, 100K a month: 3,333 requests a day, 1,500 input and 250 output tokens each, 50% of the input from cache. Data extraction: 1,667 requests a day, 2,500 input and 300 output tokens each. Bulk classification: 33,333 requests a day, 400 input and 10 output tokens each.
Each model’s monthly bill for your workload is subtracted from the baseline’s bill. The percentage is that difference divided by the baseline’s bill, and the 12-month figure is the monthly difference times twelve. The percentages under the input and output prices compare each price with the baseline’s price of the same kind.
The prompt or answer length at which two models cost the same per request. With the prompt held fixed, the cheaper model for short answers is the one with the lower effective input cost and the cheaper one for long answers is the one with the lower output price; the calculator shows where they swap and which side of that point you are on. If one model is cheaper on both input and output, there is no break-even point.
Because every line of the bill is multiplied by the same number of requests. Doubling the volume doubles every model’s bill, so the order and the percentage differences stay the same; only the amounts in money change. What moves the order is the mix of input, output and cached input in each request.
It divides your monthly budget by each model’s cost of one request in a 30-day month to get the requests a day the budget pays for. With a growth rate, each month’s bill is the previous one times one plus the rate; the planner reports the first month the bill passes the budget, looking up to 36 months ahead, and adds the first twelve bills for the 12-month total. Prices are held at the latest collection.