More and more companies are incorporating LLMs into their businesses. Additionally, there are increasing cases where companies provide LLMs that incorporate their proprietary data as a service. In such cases, the initial investment in GPUs and other hardware can be a concern.

This time, we will summarize services that allow you to easily use LLMs via APIs.

Services Providing Proprietary AI Models

OpenAI API

OpenAI offers a wide range of services, including ChatGPT and DALL-E. In particular, ChatGPT is a service adopted by many companies. It can be used not only for text but also for images, videos, and audio.

OpenAI API | OpenAI

Gemini API

Gemini is the latest AI developed by Google. There are three types of Gemini models: Nano, Pro, and Ultra. The Nano model is intended for use on smartphones. In addition to being integrated into various Google services, it can also be operated using APIs.

Google AI Gemini API | Gemma open models | Google for Developers | Google AI for Developers

Groq

Groq is an LLM developed by former Google engineers. Its key differentiator from other LLMs is its speed of response. This is achieved through the use of a proprietary processing engine called the Language Processing Unit (LPU), which is optimally tuned for Groq.

Groq is Fast AI Inference

Claude

Claude is a company founded by former executives of OpenAI. Currently, it handles text and images. At one point, it was considered superior to ChatGPT. The latest version is Claude 3.5 Sonnet, and the next version is eagerly anticipated.

Build with Claude \ Anthropic

Cohere

Cohere is an LLM specialized for enterprises. It is particularly suited for searching enterprise data, including RAG. Cohere can be deployed on the cloud, on-premises, or on a public cloud.

Login | Cohere

Mistral AI API

Mistral is a French AI startup founded by former Meta and DeepMind developers, focused on AI research and development. It has released open models like Mistral 7B and Mixtral 8x7B, and commercial models such as Mistral Small, Mistral Embed, and Mistral Large.

Mistral AI API | Mistral AI Large Language Models

Perplexity API

Perplexity combines the GPT model with search functionality to display the sources of data, which results in more reliable outputs.

Perplexity API

ALEPH ALPHA

ALEPH ALPHA is an AI service founded by former Apple developers. Based in Germany, the company aims to be the OpenAI of the EU, complying with EU data management and privacy regulations.

ALEPH ALPHA - AI for Enterprises and Governments

DeepSeek

DeepSeek is an LLM with 23.6 billion parameters. Its low cost compared to other LLM APIs is an advantage, but caution is advised as it is a Chinese AI.

DeepSeek

AI21

AI21 is an AI startup based in Israel. It offers a function to provide links to data sources. It can be used directly or via platforms like Amazon Bedrock.

AI21 Studio Documentation

APIs for Calling External LLMs

These services simplify the use of various LLM APIs. They do not have their own engines, but make it easy to try and switch between different LLMs.

Azure OpenAI Service

This service is provided by Azure. The following models are available:

• OpenAI
• Phi-3
• Llama
• Mistral AI
• Cohere
• Hugging Face
• Stability AI
• Core42
• Nixtla

Azure OpenAI Service – Advanced Language Models | Microsoft Azure

Vertex AI

Vertex AI is a fully managed integrated AI development platform provided by Google. The Model Garden showcases various AI models. While the models are primarily developed by Google, other models such as Llama 2 and Claude are also available.

• Gemini Model
• Imagen 2.0
• Chirp 2.0
• Gemma
• CodeGemma
• PaliGemma
• Llama 2
• Falcon
• BERT
• T-5 FLAN
• ViT
• EfficientNet
• Claude Model Family

Using Vertex AI with Gemini 1.5 Pro and Gemini 1.5 Flash | Google Cloud

Amazon Bedrock

Amazon Bedrock is a service provided by AWS that allows the use of various LLM models from a single endpoint. The available models are:

• AI21 Labs
• Amazon Titan
• Claude
• Cohere
• Llama
• Mistral AI
• Stability AI

Build AI Applications with Foundational Models - Amazon Bedrock - AWS

NVIDIA LLM APIs

NVIDIA provides various AI models, including Llama 2, Phi, and Mistralai, via APIs. The available models are:

1. YI-large (01-ai)
2. Sea-lion-7b-instruct (aisingapore)
3. Starcoder2 (bigcode)
4. Databricks
5. Deepseek
6. Gemma (Google)
7. Codegemma (Google)
8. Recurrentgemma-2b (Google)
9. Shieldgemma-9b (Google)
10. Granite (IBM)
11. Breeze-7b-instruct (mediatek)
12. Codellama-70b (meta)
13. Llama (meta)
14. Phi-3 (Microsoft)
15. Codestral-22b-instruct-v0.1 (mistralai)
16. Mathstral-7b-v0.1 (mistralai)
17. Mixtral (mistralai)
18. Nemotron-4 (nvidia)
19. Rakuten AI (rakuten)
20. Seallm
21. Srctic (snowflake)
22. Solar (upstage)

LLM APIs

Fireworks AI

Fireworks AI is a service launched by former Google engineers. The models available are:

1. Llama 3/3.1
2. Mixtral MoE
3. Firefunction V2
4. FireLLaVA-13B
5. Chronos Hermes 13B v2
6. CodeGemma 2B/7B
7. Code Llama
8. Stable Diffusion
9. Playground v2/v2.5

Introduction - Fireworks AI Docs

NPL Cloud

NPL Cloud offers AI APIs utilizing various models, such as Dolphin, Llama, Whisper, and Bart, which are fine-tuned for specific use cases.

Advanced AI API

Together AI

Together AI is an AI platform service launched by a former senior director of Apple, offering a platform to create open and customizable AI models.

Together AI

Replicate

Replicate offers various AI models as APIs. Charges are incurred based on the seconds used. You can run models or fine-tune them on Replicate.

Replicate — Run AI with an API

Summary

Using APIs allows for the inexpensive use of LLMs, with charges based on the number of calls or tokens. Even when creating a customized model, platforms that make it available via APIs can significantly reduce the initial investment. Consider incorporating AI into your company’s services.

Hexabase promotes AI-driven development, realizing efficient development using natural language. Check out the free alpha version of "App Modeler," an AI-driven development tool that allows requirements definition in just three minutes!