01What Is Inkling 1T Open Source Model?

Inkling 1T open source model is a new open-weight AI model from Thinking Machines. It has 975 billion total parameters with 41 billion active parameters. It uses a mixture of experts architecture and supports a context window of 1 million tokens.

The model was pre-trained on 45 trillion tokens.

Inkling is the first in a family of models of different sizes. It is also the first open source 1 trillion multimodal model that supports image, audio, and text. You can run it inside open code, and it is available for free on Hugging Face.

In this episode, Julian tested Inkling across 50 tasks. He showed what it can do, what it can't do, and how it compares to other models. The results are mixed, but the model is worth checking out because it is open source.

02Main Features of Inkling Model

Inkling has several key features that make it stand out. First, it is open source, so anyone can use it and fine-tune it. Second, it is multimodal, meaning it can handle image, audio, and text.

Third, it has a huge context window of 1 million tokens, so you can give it a lot of information at once.

  • Mixture of experts with 975 billion total parameters and 41 billion active.
  • Pre-trained on 45 trillion tokens.
  • Available for fine-tuning on Pinker.
  • Can be used inside the Inkling playground.
  • Available for free on Hugging Face.
  • Works directly inside open code, which is part of the Egentic OS.

Julian created a framework called the Ink Machine for testing. He also noted that the model is designed for one-shot generation, like creating landing pages or visual concepts. It is not meant to compete with frontier models like Fable 5, but it offers a strong open source alternative.

03How to Set Up and Start Using Inkling

Setting up Inkling is simple. Julian showed how he uses it inside open code, which is part of his agent operating system. He got an API key from Tinker, and then wired it into open code.

Here are the steps to get started:

  1. Go to tinker.thinkingmachines.ai and create an account.
  2. Grab an API key from the Tinker dashboard.
  3. Open open code (or your preferred coding environment).
  4. Add the Inkling model to your model list using the API key.
  5. Switch to the Inkling model in the model selector.
  6. Start building with prompts, such as creating a landing page.

You can also use Inkling directly on Hugging Face without an API key. The playground lets you test prompts quickly. For more advanced use, fine-tuning is available on Pinker.

Julian recommends wiring it into open code for easier testing and building.

04How to Use Inkling for Real Projects

Julian used Inkling in open code to create a landing page for an SEO agency. He also showed several examples of what the model can generate. These examples include a visual workspace that moves with the mouse, a neon clock, and a matrix wallpaper theme.

The model creates beautiful visual content with ease.

To use Inkling, you simply write a prompt. For example, you can ask it to 'create a beautiful landing page for an SEO agency' and it will generate a full page. Julian showed that the font and design are quite good.

He also built a fireworks animation that looked 'insane' and an orbit style visual.

However, Inkling struggles with 3D games. Julian tried creating a Crypt Dungeon game, a Dragon Realm open world game, and a Skyrim style game. All of them had issues.

Movement often broke, and the games were not interesting. For 3D, Inkling is not recommended.

For one-shot web design and visual art, Inkling works well. It is fast and produces nice results. But for complex tasks like physics or 3D, other models are better.

05Inkling Performance on GoldieBench and Comparisons

Julian tested Inkling on GoldieBench, which runs about 50 projects across 30 different models. Inkling scored an average of 6 out of 10 and ranked 14th. This is mediocre compared to top models like Fusion, Hermes Mixture of Agents, and GPC 5.6, which are at the top.

Inkling outperforms Nemetron 3 Ultra, but it is not as good as GLM 512, which Julian says is 'outrageously good' and the best open source model he has tried. For example, a dungeon game from GLM 512 worked, while Inkling's version totally broke. A Raycaster style game also worked beautifully with GLM 512 but not with Inkling.

Julian noted that design arena benchmarks show Inkling scoring highly, but he does not trust those benchmarks. He created GoldieBench because traditional benchmarks are not reliable. In real tests, Inkling shows clear weaknesses in 3D and physics.

Julian also mentioned that HY3 from Tencent is probably better than Inkling.

06Tips, Limits, and Mistakes to Avoid with Inkling

When using Inkling, there are several limits to keep in mind. First, it is not great at 3D games or physics. Do not expect it to create playable games.

Second, it is not on the same level as frontier open source models like GLM 512. Third, the model can produce broken outputs, as seen with the Dragon Realm example.

  • Use Inkling for one-shot web design, not complex apps.
  • Test prompts thoroughly; results can vary widely.
  • Don't rely on official benchmarks; test for yourself.
  • Fine-tuning is possible on Pinker for better results.
  • Combine with open code for easier workflow.

Julian pointed out that his results might be due to his prompting. He said he is open to being wrong. The key takeaway is to test everything yourself.

Don't trust benchmarks, even from him. Try different prompts and see what works for your use case.

07Who Inkling Is For and Why It Matters

Inkling is for developers and AI enthusiasts who want a free, open source model. It is especially good for creating visual content like landing pages and art. You can use it without paying for a commercial API, which is a big advantage.

It matters because it is an open source model with a huge context window and multimodal capabilities. This is rare in the open source world. Julian supports open source efforts, even if Inkling is not the best.

He believes open source models are worth checking out, and he is excited to see what Thinking Machines releases next.

If you value open source and want to experiment with a new model, Inkling is worth trying. You can fine-tune it and use it in your projects. However, if you need top performance, especially for 3D or coding, you might prefer GLM 512 or other models.