IBM Models
IBM watsonx.ai foundation models with public pay-as-you-go per-million token rates.
Granite 4H Small (watsonx.ai)
$0.0636 / $0.265 per 1M tokens
Llama 4 Maverick (watsonx.ai)
$0.371 / $1.484 per 1M tokens
Llama 3.3 70B (watsonx.ai)
$0.7526 / $0.7526 per 1M tokens
Mistral Small 3.1 24B (watsonx.ai)
$0.106 / $0.318 per 1M tokens
Mistral Large (watsonx.ai)
$0.636 / $1.908 per 1M tokens
GPT OSS 120B (watsonx.ai)
$0.159 / $0.636 per 1M tokens
About IBM
IBM watsonx.ai foundation models with public pay-as-you-go per-million token rates. IBM appears on CostUse as a provider of AI models with per-token pricing. This page groups the published catalog calculators so you can compare list rates, project monthly volume, and open each model with the public pricing source next to the figure.
There are 6 published models from IBM here. On a 70/30 mix, combined list rates run from $0.12402 to $1.0176 per 1 million tokens, with a median of $0.5035. The largest listed context window in this group is 1,000,000 tokens. The most recent price check in this group is 2026-07-30.
IBM model pricing table
Figures below are list prices per 1 million tokens. The 70/30 mix is a planning default—switch the mix on each model calculator if your traffic is output-heavy.
| Model | Input / 1M | Output / 1M | Mixed 70/30 | Context | Verified |
|---|---|---|---|---|---|
| Granite 4H Small (watsonx.ai) | $0.0636 | $0.265 | $0.12402 | 128,000 | |
| Mistral Small 3.1 24B (watsonx.ai) | $0.106 | $0.318 | $0.1696 | 128,000 | |
| GPT OSS 120B (watsonx.ai) | $0.159 | $0.636 | $0.3021 | 128,000 | |
| Llama 4 Maverick (watsonx.ai) | $0.371 | $1.484 | $0.7049 | 1,000,000 | |
| Llama 3.3 70B (watsonx.ai) | $0.7526 | $0.7526 | $0.7526 | 128,000 | |
| Mistral Large (watsonx.ai) | $0.636 | $1.908 | $1.0176 | 128,000 |
How to read IBM API cost
Input and output almost never share the same rate. On chat and agent workloads, completion length often dominates the bill. Use each model’s volume bands (from a handful of tokens to tens of millions) to find where a mini or flash tier beats the flagship on cost per accepted result—not only on unit price.
When the source publishes cached input, the model calculator can switch to that rate. For stable system prompts and RAG with repeated documents, cache often moves monthly spend more than swapping models inside the same family.
IBM calculators in this catalog
The cards at the top of this page open each model’s full page: list price, interactive calculator, cost profile, and FAQs. Start with the scenario closest to your product, then compare across the rest of the IBM lineup.
Frequently asked questions about IBM
How many IBM models have calculators on CostUse?
There are 6 published models from IBM, each with input and output rates (when the source splits them), volume bands, and a link to the official pricing page.
Which IBM model has the lowest list cost in this catalog?
On a 70% input / 30% output mix, Granite 4H Small (watsonx.ai) is the lowest-priced option in this group at $0.12402 per 1 million tokens. Open the model calculator to stress-test your own mix.
Do IBM prices include discounts and taxes?
No. Calculators use public list rates. Annual commits, promo credits, regions, taxes, and commercial discounts are not baked into the figure—always confirm the source linked on the model card.
How should I plan a monthly budget with the IBM API?
Pick a model, estimate input and output tokens per request, multiply by requests per day and by 30. Compare flagship tiers with cheaper ones: classification and routing often clear quality bars on smaller models at a fraction of the cost.
When is the more expensive IBM tier worth it?
Mistral Large (watsonx.ai) sits near $1.0176 per 1M tokens on a 70/30 mix, versus $0.12402 on Granite 4H Small (watsonx.ai). Use the high tier for long-horizon reasoning or fewer retries; keep the cheap tier for high volume and simple tasks.