Savings estimates · prices checked 2026-11-03
What a yes/no call costs, by model
Estimated cost per 1,000 short classification calls (500 input tokens, 5 output tokens plus estimated reasoning at each model’s default effort), and the share of calls a cheaper model must keep before a hybrid route saves 30%.
| Current model | $/M in / out | Est. per 1,000 calls | Levanto Sage 2 per 1,000 | Keep ≥ for 30% saving |
|---|---|---|---|---|
| Claude Fable 5.1 | $10 / $50 | $21.25 | $0.027 | 30% |
| GPT-6 Astra | $10 / $50 | $11.25 | $0.027 | 30% |
| Claude Opus 5.5 | $4 / $20 | $4.50 | $0.027 | 31% |
| Claude Sonnet 5.5 | $2 / $10 | $4.25 | $0.027 | 31% |
| GPT-6.1 Sol | $2 / $10 | $2.25 | $0.027 | 31% |
| Gemini 4 Argon | $2 / $10 | $2.25 | $0.027 | 31% |
| GPT-6.1 Terra | $1 / $5 | $1.13 | $0.027 | 32% |
| Gemini 3.8 Flash | $0.75 / $3.75 | $0.844 | $0.027 | 33% |
| GPT-5 mini | $0.25 / $2 | $0.375 | $0.027 | 37% |
Estimates, not bills. Reasoning-token counts are our estimates for short calls; batch and cached-input discounts are excluded. Each page shows the assumptions, effort-level costs and monthly figures.
The share a cheaper model keeps depends on your task. Measure it on your own inputs with the free cutover audit, or play with volumes in the routing calculator. Extracting fields instead? See extraction cost by model. Unsure about effort settings? What reasoning effort costs. New to this? Start with how to switch to a smaller model.
Questions
Why compare against one cheap model?
Levanto Sage 2 is the model the free audit tests today. It answers yes/no and extraction calls with a calibrated probability, which is what confidence routing needs. Same-provider cheaper models are listed on each page for comparison.
How current are the prices?
They were checked against the providers’ pricing pages on 2026-11-03 and are the same table the public prices endpoint serves. Intro prices are flagged on each page with the later price. GPT-5.6 Sol has no page: its current price is promotional (through at least Nov 21, 2026), so we will add it once the regular price is set.