Monthly and annual spend, by model.
Enter your volumes once and compare up to three alternatives against your current model. The URL updates as you type, so you can share the exact scenario.
Workload
Models
Current run-rate
Excludes cache-write premiums, taxes, negotiated discounts and tool calls, except DeepSeek Search/Fetch requests ($0.75 per 1,000 successful requests each, billed from 2026-10-31 16:00 UTC; math). Off-peak share applies only where an off-peak rate is published (DeepSeek: ~79% of weekly hours, 133 of 168, are off-peak). Batch share uses published batch rates where available; otherwise standard. Set typical prompt size above a model's long-context threshold to apply that tier.
Scenario comparison
| Model | Monthly | Annual | Annual savings vs current | Δ % | Source · checked |
|---|
Cost per task, at a glance
A useful sanity check: what one unit of work costs at standard rates.
| Model | Support reply 3K in / 500 out | RAG answer 12K in / 700 out | Code review 40K in / 2K out | Per 1K support replies |
|---|---|---|---|---|
| gpt-6-luna · OpenAI | $0.00055 | $0.00155 | $0.0050 | $0.55 |
| Claude Haiku 5.5 · Anthropic | $0.00055 | $0.00155 | $0.0050 | $0.55 |
| gpt-5.6-luna · OpenAI | $0.00120 | $0.00324 | $0.0104 | $1.20 |
| deepseek-flash (V4.1 Flash) · DeepSeek | $0.00150 | $0.00444 | $0.0144 | $1.50 |
| Gemini 3.1 Flash-Lite · Google | $0.00150 | $0.00405 | $0.0130 | $1.50 |
| Gemini 3.5 Flash-Lite · Google | $0.00215 | $0.00535 | $0.0170 | $2.15 |
| Gemini 3 Flash Preview · Google | $0.00300 | $0.00810 | $0.0260 | $3.00 |
| Gemini 3.8 Flash · Google | $0.00413 | $0.01162 | $0.0375 | $4.13 |
| Gemini 3.7 Flash · Google | $0.00413 | $0.01162 | $0.0375 | $4.13 |
| Gemini 3.6 Flash · Google | $0.00413 | $0.01162 | $0.0375 | $4.13 |
| deepseek-v4-pro · DeepSeek | $0.00594 | $0.01861 | $0.0607 | $5.94 |
| gpt-6.1-terra · OpenAI | $0.00550 | $0.01550 | $0.0500 | $5.50 |
Track this budget automatically
Ratecard Pro sends an alert when a rate you depend on changes, and when a forecast crosses your budget. See Pro.