Granite 4.2 8b API Pricing

These are the tracked rates, not a live quote. Hosted-provider choice, region, long-context tiers, caching, batch discounts, and temporary promotions can change the actual bill. Open-weight availability does not mean hosted inference is free. Workload examples assume the listed standard input and output rates, with no cache hits or discounts.

Publisher-reported pricing notes

• The model may produce inaccurate, biased, or unsafe responses, and IBM recommends deployment alongside additional safety controls.

Official release and model references

These references document the model and its release; not every reference is a pricing table. Confirm the applicable endpoint and billing tier with your inference provider before purchase.

• www.ibm.com/granite/docs/models/granite4-2
• huggingface.co/ibm-granite/granite-4.2-8b
• huggingface.co/blog/ibm-granite/granite-4-2
• artificialanalysis.ai/models/granite-4-2-8b

Computed straight from the list prices above — token counts are illustrative workload sizes, not measurements of Granite 4.2 8b.

In AI IQ's scoring, Granite 4.2 8b's list price is adjusted by a token-usage multiplier of 0.81 (measured from real benchmark runs), giving an effective cost of $0.2511 per 1M input + output tokens. That places it #11 of 146 models on the effective-cost ranking (lower is cheaper). Some models spend far more tokens than others on the same task, so effective cost compares what a unit of work really costs. Method details are on the methodology page; see all models on the cost charts and the falling cost of intelligence over time.

Models with a lower effective cost that score within a few IQ points of Granite 4.2 8b (IQ 95), or better.