I tried to parse hand-drawn ER diagrams in the past and did not have much success with any model, frontier or otherwise. If you have annotated data, I'd recommend finetuning a recent (dense) VLM, but don't expect 100% accuracy. https://unsloth.ai/docs/basics/vision-fine-tuning

    LLM issues tool call to read high res image ->
    harness sends high res image to server ->
    server downsizes it to 800x800 (blurry) ->
    LLM issues bash command (e.g. `convert`) to crop a small subimage (e.g. 600x600) from the high res image ->
    LLM issues tool call to read subimage ->
    harness sends subimage to server ->
    server does not resize the subimage because it is small already, so it is not blurry when finally ingested by the LLM

    Nevertheless, as a component, we will undoubtedly implement multimodal support — and we are already doing so. We plan to develop relevant models, ensuring that versions like V4 and subsequent iterations will natively support multimodal functionality.

    Achieving excellence in AI training does not require a global model or even multimodal approaches—by narrowing the scope of AI training and eliminating multimodality, certain tasks may remain unachievable without compromising the algorithm's validity.

    Multimodal approaches ultimately need to be implemented.

I am using a stripped-down minimal version of it which I uploaded here, since I am not a fan of huge dependency trees: https://github.com/99991/simple-pp-doclayoutv3

Another recent model for this task is Unlimited-OCR: https://github.com/baidu/Unlimited-OCR

"Our billion-dollar infrastructure crumbles under a tremendous flood of 50 PRs per second" would just sound embarrassing.

The current benchmark suites that frontier AI labs use are probably a good fit, e.g.

https://z.ai/blog/glm-5.3#:~:text=Performance%20across%20com...

https://www.kimi.ai/ai-models/kimi-k3#:~:text=Performance%20...

https://www.anthropic.com/news/claude-opus-5

https://openai.com/index/gpt-5-6/

But guessing from your current benchmarks, I assume that you are severely compute-constrained. What is your time budget?

I could of course download and test myself, but that would take days with my internet connection.

Also, TrustedRouter's website is full of slop, which does not inspire much confidence.

    size_t i{0};

    size_t i = 0;

    size_t i{0.0};

    double d = 0.0;
    size_t i{d};

In C++, you can get the same effect without the unusual syntax by passing -Wfloat-conversion to gcc/clang, but not sure how to do that with CUDA:

The former has a button to dismiss the dialog. Maybe you clicked on it by accident, or maybe it does not work right.

    > We plan to raise the overall pricing for DeepSeek API services in the near future, with a significant increase expected. Please plan your usage accordingly. The specific pricing plan will be subject to official notice.

And there is also the reason for why I keep all this trash around in my house instead of throwing it away. The odds that I'll ever need it are pretty slim, but if I should need it, it would be a hassle to reacquire, and it costs me almost nothing to keep it around.

https://huggingface.co/nvidia/GLM-5.2-NVFP4

Interesting. Why is that? I would have expected the opposite, since larger models have to try less hard to fit the training data. Or maybe this leaves more parameters with random initialization, resulting in higher entropy for larger models?

You can!

> AFM 3 Core Advanced makes routing decisions per prompt. A lightweight, dense block selects a fixed set of experts during initial processing, periodically reselecting them during generation.

https://machinelearning.apple.com/research/introducing-third...

* Sparse Experts: 1481.4 GB

* Dense Experts: 1.9 GB

* Self-Attention: 72.4 GB

* LLM Head: 2.4 GB

* Embeddings: 2.4 GB

* Vision Encoder: 0.35 GB (surprisingly small)

plus some miscellaneous parameters.

Most importantly, we now know that the model has 104B active parameters, which is quite a lot and will make it difficult to self-host efficiently.