nele.ai – The billing system and how to use your AI volume efficiently
nele.ai is an application that connects various AI models, text, code, and spreadsheets, ensuring that no personal data is sent to the AI models and no sensitive company data is used for training purposes.
Models for generating audio, video, and image files will also be available in the future.
To integrate the various AI models and billing systems within nele.ai, a comprehensive billing unit (credits) has been developed.
To ensure investment security when working with AI, nele.ai uses a monthly data volume (AI volume) subscription, which means there are no additional costs per user. This allows nele.ai to be made available to every employee while keeping cost risks limited by the defined AI volume.
The AI volume can be adjusted at any time if needed.
To help you get the most out of your AI volume, we have listed a few important points to keep in mind:
1. Choose the right AI model for your task!
Currently, various AI models are available for your tasks. Each AI model has an individual consumption factor due to its varying parameter size and processing requirements.
Here are the most important differences:
OpenAI
- GPT 4o 128k
- Context size: 128,000 tokens
- Consumption factor: 0.25
- Description: Most advanced model with 128K context
- GPT 4 Turbo 128k
- Context size: 128,000 tokens
- Consumption factor: 0.5
- Description: New model with 128K context
- GPT 4 8k
- Context size: 8,192 tokens
- Consumption factor: 1
- Description: High-quality responses
- GPT 3.5 Turbo 4k
- Context size: 4,096 tokens
- Consumption factor: 0.04
- Description: Fast and cost-effective
- GPT 3.5 Turbo 16k
- Context size: 16,385 tokens
- Consumption factor: 0.08
- Description: Allows for more words than GPT 3.5 Turbo 4k
- DALL·E 3
- Context size: 4,000 tokens
- Consumption factor: 20-60 credits/image
- Description: High-quality image generation
Anthropic
- Claude Opus 200k
- Context size: 200,000 tokens
- Consumption factor: 1
- Description: Best Anthropic model with 200K context
- Claude Sonnet 200k
- Context size: 200,000 tokens
- Consumption factor: 0.2
- Description: Balanced Anthropic model with 200K context
- Claude Haiku 200k
- Context size: 200,000 tokens
- Consumption factor: 0.02
- Description: Fastest Anthropic model with 200K context
Azure (Microsoft)
- GPT 4 Turbo 128k
- Context size: 128,000 tokens
- Consumption factor: 0.5
- Description: New model with 128K context
- GPT 4 8k
- Context size: 4,096 tokens
- Consumption factor: 1
- Description: High-quality responses
- GPT 4 32k
- Context size: 32,768 tokens
- Consumption factor: 2
- Description: Allows for more words than GPT 4 8k
- GPT 3.5 Turbo 4k
- Context size: 4,096 tokens
- Consumption factor: 0.04
- Description: Fast and cost-effective
- DALL·E 3
- Context size: 4,000 tokens
- Consumption factor: 20-60 credits/image
- Description: High-quality image generation
OpenAI
- GPT 4o 128k: This advanced model has a context size of 128,000 tokens and a consumption factor of 0.25.
- GPT 4 Turbo 128k: A new model with a context size of 128,000 tokens and a consumption factor of 0.5.
- GPT 4 8k: This model provides high-quality responses with a context size of 8,192 tokens and a consumption factor of 1.
- GPT 3.5 Turbo 4k: With a context size of 4,096 tokens and a consumption factor of 0.04, this model is fast and cost-effective.
- GPT 3.5 Turbo 16k: This model allows for more words than GPT 3.5 Turbo 4k, with a context size of 16,385 tokens and a consumption factor of 0.08.
- DALL·E 3: This model offers high-quality image generation with a context size of 4,000 tokens and a consumption factor of 20-60 credits per image.
Anthropic
- Claude Opus 200k: The best Anthropic model has a context size of 200,000 tokens and a consumption factor of 1.
- Claude Sonnet 200k: A balanced Anthropic model with a context size of 200,000 tokens and a consumption factor of 0.2.
- Claude Haiku 200k: The fastest Anthropic model has a context size of 200,000 tokens and a consumption factor of 0.02.
Azure (Microsoft)
- GPT 4 Turbo 128k: A new model from Microsoft with a context size of 128,000 tokens and a consumption factor of 0.5.
- GPT 4 8k: This model offers high-quality responses from Microsoft with a context size of 4,096 tokens and a consumption factor of 1.
- GPT 4 32k: This model allows for more words than GPT 4 8k, with a context size of 32,768 tokens and a consumption factor of 2.
- GPT 3.5 Turbo 4k: A fast and cost-effective model from Microsoft with a context size of 4,096 tokens and a consumption factor of 0.04.
- DALL·E 3: This model from Microsoft offers high-quality image generation with a context size of 4,000 tokens and a consumption factor of 20-60 credits per image.
2. When should you use which model?
Use a cost-effective and efficient model for simpler tasks, such as answering emails, writing speeches, drafting employment references, etc.
Use an extensive and sophisticated model for complex tasks, such as creating concepts, contracts, business plans, or scientific summaries, etc.
Important note:
Comparing the effort and costs without using AI can help with decision-making.
3. Keep an eye on your input and output volume!
A major advantage of working with generative chatbots is the chat function. This allows information to be continuously expanded and improved within the chat history.
However, please note that both sent (input) and received words (output) count toward the total AI consumption. This means that every time you send a message, the entire chat history is sent. Therefore, total AI volume consumption can increase faster than expected.
Tip:
It is advisable to start a new chat when changing the chat context or to divide the chat into smaller, logical units.
4. Optimize your prompts!
Your prompts should be clear and specific. Vague prompts can lead to longer and less accurate responses.
Example prompt:
Write a LinkedIn post from a user's perspective, limited to 300 characters with relevant hashtags and a few emojis. The text is: {{text}}
This prompt includes the persona/character (user), the character count (300), and information regarding hashtags and emojis.
!!The more precise the prompt, the better the output!!
Tip:
It can be helpful to ask GPT-3.5 open-ended questions to gain inspiration for specific topics. You can then have GPT-4 process these professionally.
5. Test your prompts with small text samples!
Start by having your prompts process small amounts of text and adjust them according to the desired results. Once the samples are satisfactory, you can use the prompts for larger volumes.
6. Avoid additional explanations if possible!
Depending on the AI model selected, nele.ai often provides explanations for its answers. However, if you are already familiar with the topic and do not need explanations, you can phrase your prompt to skip them. This reduces the amount of text and saves time during editing.
Simply add the phrase "no further explanation required" to your prompt.
We wish you every success in your work with nele.ai.
