Frequently asked questions
- How does token being counted?
A token is a single word, number, or symbol. LLMs (Large Language Models) use tokens to represent text. For example, the sentence "I love you" would be represented as the following tokens: "I", "love", "you". Some long words are split into multiple tokens. For example, the word "university" is split into "un", "##iv", "##ers", "##ity", so they are represented as 4 tokens. Usually, 750 English words are about 1000 tokens, and 1000 Chinese characters are about 2500 tokens.
- How does points being deducted?
The minimal billable unit is 50 tokens, which means that everything will be rounding up to 50 tokens and 1 integer point. That means, if a model charges 10 points per 500 tokens, and your consumption is 50 tokens, you will only be charged 1 point. That also means, due to the price of the GPT-3.5, you will be charged 1 point in both a consumption of 50 tokens and a consumption of 500 tokens. Both prompt (the message you sent) and response (the message you received) will be counted in the token consumption.
- What model should I use?
The model you should use depends on your purpose and budget. Generally, GPT-4 is outperformed by GPT-3.5 in terms of accuracy and creativity. However, GPT-3.5 is also great at generating creative text, answer basic questions, and is much cheaper than GPT-4. If you need a model that is highly accurate for reasoning (e.g. math), GPT-4 is the better choice. If you are on a budget and only need basic logics and text generation, GPT-3.5 may be a better option. By default, you will be using GPT-3.5 because it will be sufficient and economical for most of the use cases.
- Can I use Cockpit totally free?
Yes, you can use Cockpit totally free if you have access to it. Current registration flow is invite-only. There will be some point limits per month for free users, but it is more than enough for most of the use cases if you are using GPT-3.5 rather than GPT-4.
- How can I pay for the plan?
Right now, we are not selling any subscription publicly. However, if you already have the access to Cockpit and you want to pay for the plan (for higher quota or supporting development), please contact us via email or WeChat.
- Why am I charged even if my question to GPT-3.5 is short?
We have temporary discount for GPT-3.5, which covers the cost of the first 500 tokens. However, if you are including previous messages (via max-depth configuration), you might still exceed the 500 tokens limit because all previous messages are counted as well. So please be aware when enjoying the temporary discount. This discount is effective since Apr 5, 2023 for supporting academic exploration, and subject to change in the future.
- What if I have more questions?
Contact us via email ([email protected]) or WeChat. We will be happy to help you.