It seems like every day on X there is a post about a new model overtaking Claude or GPT. The belle of the ball this week was DeepSeek V4.1 Flash which is ~97% cheaper than GPT 5.6 Sol. Models like GLM 5.3 and Kimi K3 have also entered the arena as work horses (or work oxen?) as of late.

But where and how do you try all these shiny new models? And why?

The why is easy. Price and speed. Models like Deepseek can be your implementation power houses, saving you tokens. A nice workflow I have seen is to use Claude or GPT to come up with a plan, then point Deepseek to that plan to do the raw implementation - faster and cheaper.

Where can you try them? Well you need a harness. And an API. And that's where Oh-My-Pi and Oxen.ai come in.

Oh-My-Pi 🤝 Oxen.ai

Oh-My-Pi or omp is an incredible and fun open source harness you can run on the command line. If you want a user interface - this post is not for you. If you are a Claude Code or Codex fan from the terminal, welcome to the light side.

Here is the github repository for the harness itself:

To install it, simply use:

curl -fsSL https://omp.sh/install | sh

Once you have it installed, you will need to connect it to some models. Oxen.ai has access to all the latest closed and open source models via the same API. The APIs are all OpenAI compatible, so they are easy to swap into any tool that already supports OpenAI or OpenRouter.

Add the following provider to ~/.omp/agent/models.yml. If the file already contains a providers mapping, merge oxenai into it. Replace YOUR_API_KEY with your Oxen API key.

providers:
  oxenai:
    baseUrl: https://hub.oxen.ai/api/ai
    api: openai-completions
    apiKey: YOUR_API_KEY
    models:
      - id: deepseek-v4-1-flash
        name: DeepSeek V4.1 Flash
        reasoning: true
        input: [text, image]
        contextWindow: 1048576
        maxTokens: 384000
        cost:
          input: 0.39
          output: 1.56
          cacheRead: 0
          cacheWrite: 0

      - id: deepseek-v4-pro
        name: DeepSeek V4 Pro
        reasoning: true
        input: [text]
        contextWindow: 1048576
        maxTokens: 384000
        cost:
          input: 1.15
          output: 2.25
          cacheRead: 0
          cacheWrite: 0

      - id: kimi-k3
        name: Kimi K3
        reasoning: true
        input: [text, image]
        contextWindow: 1048576
        maxTokens: 1048576
        cost:
          input: 3
          output: 15
          cacheRead: 0
          cacheWrite: 0

      - id: zai-org-glm-5-3
        name: GLM 5.3
        reasoning: true
        input: [text]
        contextWindow: 1048576
        maxTokens: 131072
        cost:
          input: 1.82
          output: 5.72
          cacheRead: 0
          cacheWrite: 0

Once you have the initial pattern down, and one or two models integrated, you can always ask just ask any of the models to update the ~/.omp/agent/models.yml file and add new models itself by getting the information from the Oxen API:

curl -fsSL https://hub.oxen.ai/api/ai/models

Oxen returns token prices per token. Multiply by 1,000,000 for the config’s cost.input and cost.output values. For example, DeepSeek’s input price of 0.00000009 becomes 0.09 dollars per million tokens.

To see what models you have available via oxenai you can then run:

omp models oxenai

To run a specific model, simply run:

omp --model oxenai/deepseek-v4-1-flash

And you are off to the races! Happy hacking 🤓 🐂

PS: If you want free credits...just email us your use case at hello@oxen.ai. We're a friendly bunch, come say hello.