Choose an LLM Service

Three kinds of LLM Services

Essentially, LLM services can be divided into three categories:

• Those that operate the models directly, such as OpenAI.

  For these providers, simply search for the name of a popular large language model along with “open platform” to find the portal for obtaining an API key.

  These services are the most reliable. In fact, they have access to all users' chat logs and may use them to train future models; however, these large companies generally have clear privacy and data processing policies, so you need not worry if your rights are violated—you have a vast user base standing with you.

• Authorized services that run models on private clouds, such as Google Vertex AI, AWS Bedrock and Microsoft Azure.

  These services are also generally reliable. They are primarily chosen for their reliability and professionalism, and many enterprises start their LLM adoption journey here.

• Third-party service providers, such as OpenRouter, OpenCode, and Ollama Cloud.

  While these services sometimes run their own models, the vast majority obtain models from other LLM providers and forward them to you.

  The advantage of this approach is that you can access models from various providers in a single, unified platform. The downside is that you must place a high degree of trust in these providers to ensure they do not misuse your conversations with the large language models.

Choose the model that's right for you

Generally speaking, people first decide which model they want to use, and then look for the corresponding LLM service.

Google's Gemini model performs well in terms of general knowledge. If you need a knowledgeable model to help you solve everyday problems or for educational purposes, this is the one to choose.

OpenAI's well-known ChatGPT series is more general-purpose. Although recent models have faced criticism regarding writing and everyday conversation, the latest GPT-5.5 version is showing some improvements. If you're accustomed to using ChatGPT or their programming models, you can use it.

Anthropic's Claude series is renowned for its powerful programming capabilities. In fact, this also translates to strong logical reasoning and research capabilities. However, it's important to note that, according to their terms of service, you cannot use Claude Pro/Max subscriptions within Optic, and API key-based billing is quite expensive.

DeepSeek is an AI lab based in China. If the models mentioned above are considered to be in the “zero tier,” DeepSeek's models rank in the “first tier.” The key point is that, thanks to their continuous optimization of hardware performance, their models are quite affordable—even cheaper than other open-source models from China—while maintaining solid capabilities, making them worth a try.

Additionally, there are models from China such as GLM-5.1, Kimi K2.6, Qwen, Mimo, and MiniMax. If you follow these developments, you might want to give them a try.

Then select the appropriate LLM service

In Optic, we've integrated multiple LLM services. You can simply select a model from the model selector, open the corresponding web platform, generate an API key, and enter it.

If you want to know which LLM services offer a particular model, just search online—it's that easy.