Blog

GPT-6.2 gives Astra lower input and output prices

Practise AIF-C01 model selection by calculating the cost of the same hypothetical request on GPT-6.2 Astra and Sol.

· 2 min read

OpenAI’s November 17 release of GPT-6.2 Astra and Sol cut Astra’s uncached input and output rates by 20% compared with GPT-6 Astra. GPT-6.2 Sol kept the rates of GPT-6.1 Sol. The change offers a worked example for people studying AWS Certified AI Practitioner (AIF-C01). The AWS exam guide includes cost, response time and input and output length among the criteria for choosing a foundation model.

Price the same request on both models

In the OpenAI pricing table, Standard short-context rates for a million tokens of uncached input and output are $8 and $40 for GPT-6.2 Astra, and $2 and $10 for GPT-6.2 Sol. Tokens are the pieces of data the model processes or generates. The table lists separate rates for other request sizes and processing modes.

Suppose one request uses 100,000 uncached input tokens and 10,000 output tokens. On Astra, the input costs $0.80 and the output costs $0.40, for $1.20 total. On Sol, the same usage costs $0.20 for input and $0.10 for output, for $0.30 total. Both calculations use the published Standard rates and leave out tools and other charges. If you call the model from software, use the developer pricing table. OpenAI’s billing guide says the application programming interface (API) and ChatGPT have separate billing systems.

Choose a model for an accepted result

Imagine each attempt in the example uses the same number of tokens. One accepted Astra answer and four Sol attempts would each cost $1.20 in model tokens. That is a break-even calculation for the exercise. To choose a model for a real task, decide what an acceptable answer looks like, then measure how often each model meets it, how long the task takes and how many tokens each attempt uses.

Try the same calculation with a short question that requires a long answer. Ask whether cost, response time or the quality of the answer would limit your choice first. The AIF-C01 study guide covers how those requirements shape foundation model decisions. The OpenAI rates here apply to this worked example. Use the price sheet for the service named in any exam scenario.

Read the AIF-C01 study guide

Sources