[
  {
    "category": "fixes",
    "description": "**read_file** now correctly passes embedded base64 images to the model as vision tokens instead of truncated fragments.",
    "breaking_change": false
  },
  {
    "category": "features",
    "description": "**image_gen** now uses the higher-quality grok-imagine-image-quality model.",
    "breaking_change": false
  }
]