Models like Nemotron 3 Ultra are very slow

Description

Whenever I go on OpenCode to do my daily developing, I am happy with the idea and view, but the model speed makes me upset.

Main Issue

I select nvidianemotron-3-ultra-550b-a55b via the NVIDIA endpoint and begin my work. Usually, my first message does not go far. Maybe it does 5 tool calls and then quits, which is a big issue because I may be doing IRL tasks while I believe it is working. Or, when I give another prompt and it works for a while, it can take 45+ minutes even for small changes. Usually, when working, it is very slow, even though messages come through fast.

How I Know It Is OpenCode and Not the Model

When using different applications (not going to specify unless asked due to advertising) (agent development environments) that are almost identical to OpenCode, I use the same model and provider. It is just much faster, with an average of 91.1% less time (4 minutes) for minor changes.

How It Affects Me and Others

When I need to implement something quickly, this really throws off my day and actually makes me want to no longer use the app. This is likely an issue for many other users as well.

Thank You

I ask that you please look into and fix this as soon as possible so I can have a much better experience.

Plugins

N/A

OpenCode version

1.18.16 (Happens throughout all versions)

Steps to reproduce

1. Make the current code use the same system prompting/effect while significantly lowering the amount of input sent to the model.
2. Make tools and other features easier and faster for the AI to access and use.
3. Cut down unnecessary context, tool output, and system information being sent with each request.
4. Improve the agent loop so there is less overhead between tool calls and model responses.
5. Prevent the agent from stopping after only a few tool calls when it still has work left to do.
6. Improve context handling for larger projects so the model does not have to process the entire project repeatedly.
7. Reduce the amount of processing OpenCode does around each request so the same model can perform closer to its speed in other agent environments.

Screenshot and/or share link

This issue does not have many screenshots due to the confidentiality of my chats. I apologize for the lack of media.

Operating System

Microsoft Windows 11 Pro | Version 10.0.26200 | Build 26200 | 64-bit

Terminal

Windows Terminal,PowerShell,GIT Bash