We are living in the age of abundance of amazing LLMs. It is not just ChatGPT or Claude. There are new age of tools from startups as well as other companies which include OSS models (where weights are open-sourced)

Many companies have closed-source frontier models that are great at certain tasks.

For instance,

1. OpenAI's gpt-4 and the new o1 series are great at reasoning

2. Claude's Sonnet is great at Coding tasks.

3. Google's Gemini Flash models are great for blazingly fast and long content windows respectively. It is great for Academia and research purposes.

Not to mention the OSS AI models that are also taking by storm:

1. Llama 3.2 models are great at creating marketing content.

2. Cohere’s c4ai-command-r-plus are great for RAG applications.

3. Deepseek models is one of the cheapest model that rivals Open AI's o1 and Claude's Sonnet in reasoning as well as code generation respectively.

All this makes it imperative to have a tool to play around models and especially route them a a proxy level.

This is where Open Router comes in. It provides a proxy layer to call any model whether it be OSS or Closed source.

Open Router also has rankings page on usage as well as the quality metrics for different use-cases such as programming, legal, translation etc.

Credits

You can add credits to OpenRouter who has access to all the models available

For UI, you have many options to use OpenRouter

Option 1: Open Router Chat

You can use the OpenRouter chat which allows to select any model which is available. Nowadays, Deepseek is the goto model for its cost and comparable benchmarks with frontier models such as OpenAI o1, Claude Sonnet and Google’s Gemini.

You can also do web search for most models which helps in getting latest information after the training period.

You can tweak parameters such as to make the model more custom to your needs. Parameters that can be tweak include:

1. Max Tokens

2. Chat Memory

3. Temperature: How creative you want the LLM response

4. Top P

5. Top K

6. Penalty: Frequency, Presence and Repetition

7. Min P

8. Top A

Option 2: Anything LLM

There are chat GUIs that can be installed locally such as OpenWebUI, AnythingLLM, LMStudio etc

AnythingLLM provides a configuration to set OpenRouter along with Model selection:

The output is generated the using the model selected such as Open AI gpt-40 or o1 etc, Claude Sonnet, Deepseek etc.

Resources

1. Open Router

2. AnythingLLM