With per-token pricing, the model you pick is the single biggest lever on your bill. Running a frontier model on a simple classification job can cost 10–90× more than a smaller model that handles it just as well. Here's a real price comparison and where the cheaper option is usually enough.
All prices are USD per 1M tokens (input / output), list prices — the savings below come from model choice, not a discount.
The price spread
| Model | Input | Output |
|---|---|---|
| Claude Opus 4.8 | $5.00 | $25.00 |
| Claude Sonnet 4.6 | $3.00 | $15.00 |
| Claude Haiku 4.5 | $1.00 | $5.00 |
| GPT-4o | $2.50 | $10.00 |
| GPT-4o mini | $0.15 | $0.60 |
| Gemini 2.5 Pro | $1.25 | $10.00 |
| Gemini 2.5 Flash | $0.30 | $2.50 |
| DeepSeek V4 Flash | $0.10 | $0.20 |
Where a cheaper model is usually enough
Simple tasks (classification, extraction, rewriting, batch jobs) → GPT-4o mini instead of GPT-4o
$0.15 vs $2.50 input, $0.60 vs $10 output — ~94% cheaper. For structured, well-defined tasks the mini model rarely loses quality.
Everyday coding & general work → Claude Sonnet 4.6 instead of Opus 4.8
$3 vs $5 input, $15 vs $25 output — 40% cheaper. Save Opus for genuinely hard reasoning.
High-concurrency lightweight calls → Claude Haiku 4.5 instead of Sonnet 4.6
$1 vs $3 input, $5 vs $15 output — ~67% cheaper.
Long-document summarization & retrieval QA → Gemini 2.5 Flash instead of 2.5 Pro
$0.30 vs $1.25 input, $2.50 vs $10 output — ~75% cheaper.
Chinese-language work & high-volume throughput → DeepSeek V4 Flash instead of GPT-4o
$0.10 vs $2.50 input, $0.20 vs $10 output — 96–98% cheaper.
The catch: routing by hand is tedious
Knowing the cheaper model exists is one thing; actually picking it per request — and remembering to switch back for the hard prompts — is friction nobody keeps up with.
That's the problem Modelis is built around: it's an OpenAI-compatible gateway that classifies each request by difficulty and routes it to a fitted model automatically, then bills a flat per-call price so your cost stays predictable regardless of which model answered. Every response carries an X-Modelis-Routed-Model header so you can see exactly which model handled it.
Free tier — get a key.
Top comments (0)