OpenAI charges for API access using a consumption-based model measured in tokens. Costs depend heavily on the specific model and the type of data being processed, with input tokens generally priced much lower than output tokens. For instance, the entry-level gpt-5.4-nano costs $0.20 per 1M input tokens, while the high-performance gpt-5.5-pro costs $30 per 1M input tokens.
Output costs scale with model complexity. While gpt-5.6-luna provides a cost-effective output rate of $6 per 1M tokens, the gpt-5.5-pro model reaches $180 per 1M tokens. Specialized tasks like audio processing or image input carry their own specific rates, such as $8 per 1M tokens for image data.
How do context length and caching affect the bill?
Price points change based on how the model handles data. The gpt-5.6-sol model illustrates this variability: standard short context input costs $5 per 1M tokens, but this drops to $0.50 if the input is cached. Conversely, if the task requires long context processing, the rate for the same model increases to $10 per 1M input tokens and $45 per 1M output tokens.
What are the costs for audio and specialized research?
Multimedia and deep reasoning tasks use different pricing tiers. Real-time audio via gpt-realtime-2.1 is one of the more expensive services at $32 for input and $64 for output per 1M tokens. For high-volume research tasks, o3-deep-research offers a batch input rate of $5 per 1M tokens, providing a structured cost for intensive data processing.
To determine if these token costs are a sound investment, organizations must measure the cost per task against the human time saved. A model like gpt-5.6-terra, costing $2.50 per 1M input tokens, may be more capital-efficient than a pro-tier model if the output quality meets the required threshold. NetLift allows finance teams to track this unit economics by comparing API spend directly against labor hours reclaimed, ensuring AI adoption delivers a measurable net return.
Sources
All pricing on this page comes from official vendor pages: