TL;DR
- Google is rolling out Gemini Pro for developers and enterprises to build off of.
- It will be free to use until Google rolls out general availability.
- Gemini Pro will be cheaper to use than GPT-4 when the free period ends.
After announcing its new large language model (LLM) Gemini, Google started rolling the AI out to its various products. In addition to being available in its products, the tech giant is now allowing developers and enterprises to start tinkering with Gemini.
Today, Google announced that developers and enterprises can start using the Pro version of Gemini to build their own use cases. Gemini Pro will reportedly be able to take in and output text, while Gemini Pro Vision will be able to accept text and images but will only output text. It will also support 38 languages including English, Arabic, Dutch, French, Japanese, Spanish, and more.
Google says this release of Gemini will come with a 32K context window, but larger context windows will come in the future. The company also adds that Gemini Pro and Pro Vision are free to use through Google AI Studio and that Python, Android (Kotlin), Node.js, Swift, and JavaScript are all supported. However, it will stop being free once general availability is rolled out.
On Gemini Pro’s pricing page, there are two plans: free for everyone and pay-as-you-go. The free version is limited to 60 queries per minute (QPM), while the other option starts at 60 QPM. When you reach the pay-as-you-go threshold, Google will charge an input fee of $0.00025 per 1,000 characters or $0.0025 per image and an output fee of $0.0005 per 1,000 characters. If our math is correct, this means Gemini Pro is a good deal cheaper than GPT-4.
In comparison, OpenAI uses a token system, charging an input cost of $0.03 per 1,000 tokens and $0.06 per 1,000 tokens for output. OpenAI states that one token is about four characters, so 1,000 characters would be about 250 tokens. Which means 1,000 tokens is about 4,000 characters. Using OpenAI’s standard, this means Gemini Pro charges about $0.001/1K tokens for inputs and $0.002/1K tokens for outputs. In the table below, you can see that Gemini Pro is actually priced similarly to GPT-3 Turbo.
| Assuming 4 characters = 1 token | Gemini Pro | GPT-3 Turbo | GPT-4 | GPT-4 Turbo |
|---|---|---|---|---|
| Assuming 4 characters = 1 token
Input cost per 1000 char |
Gemini Pro
0.00025 |
GPT-3 Turbo
0.00025 |
GPT-4
0.0075 |
GPT-4 Turbo
0.0025 |
| Assuming 4 characters = 1 token
Output cost per 1000 char |
Gemini Pro
0.0005 |
GPT-3 Turbo
0.0005 |
GPT-4
0.015 |
GPT-4 Turbo
0.0075 |
| Assuming 4 characters = 1 token | Gemini Pro | GPT-3 Turbo | GPT-4 | GPT-4 Turbo |
| Assuming 4 characters = 1 token
Input cost per 1000 tokens |
Gemini Pro
0.001 |
GPT-3 Turbo
0.001 |
GPT-4
0.03 |
GPT-4 Turbo
0.01 |
| Assuming 4 characters = 1 token
Output cost per 1000 tokens |
Gemini Pro
0.002 |
GPT-3 Turbo
0.002 |
GPT-4
0.06 |
GPT-4 Turbo
0.03 |
| Assuming 4 characters = 1 token | Gemini Pro | GPT-3 Turbo | GPT-4 | GPT-4 Turbo |
| Assuming 4 characters = 1 token
Input price x vs Gemini Pro |
Gemini Pro | GPT-3 Turbo
1 |
GPT-4
30 |
GPT-4 Turbo
10 |
| Assuming 4 characters = 1 token
Output price x vs Gemini Pro |
Gemini Pro | GPT-3 Turbo
1 |
GPT-4
30 |
GPT-4 Turbo
15 |
It would be fair to point out that Google hasn’t released the pricing for Gemini Ultra yet — Google’s most powerful version of Gemini. However, Google states Ultra will launch sometime early next year.
