Massive garbage output (repeated characters) on long responses

Description

When generating a long response that includes code analysis and multiple findings, the model output suddenly degrades into a massive block of repeated characters (thousands of 'X' characters), corrupting the entire response.

Steps to Reproduce

1. Ask opencode to analyze a large C file (~1300 lines) for bugs and inconsistencies
2. The model begins listing findings normally
3. At some point, the output devolves into repeated garbage characters

Expected Behavior

The response should contain a clean, readable summary of findings without any rendering artifacts.

Actual Behavior

The response contained a massive block of approximately 50,000+ repeated 'X' characters appended to the end of the output, making the response unusable.

Impact

This completely ruins the response and requires the user to ask the model to repeat itself. It appears to be a token generation or output rendering issue in the model inference pipeline.

Environment

• Model: opencode/mimo-v2.5-free
• Date: 2026-08-26