ASS v5.0 is an elusive idea floating around for quite a while. It is expected to be triggered by a version header, and would change parsing to allow new tags to be used, and old VSFilter-emulation quirks to be skipped. The current draft of the specification is available on this wiki.

• Completely spec out everything before implementation
• Native, fast, easy, flexible gradient support
  • Simple linear (with direction specifier) and radial (with center specifier) with multiple stops, similar to CSS3
• Remove all VSFilter-compat hacks
• Corner pinning
• Vertical spacing tag (\fspy)
• Escape character using backslashes (e.g. \\N, \{)
  • Unicode escaping, e.g. \u2764 to encode ❤
• Well-defined semantics for \fs and font hinting
  • \fs defines font height, not line height
• Well-defined semantics for colors
• Well-defined and sane behavior and ordering of margins, rotations, shears etc.
• No \be
• Sane \blur
• Clip mask blur
• Sane layering (sane from a user's perspective; not merely easy to implement)
• Differentiate between sign typesetting and normal subtitles (beware: animated normal subtitles)
  • color matching: signs should match video, subtitles screen?
  • font hinting: probably no hinting for signs, subtitles possibly hinted?
  • rescaling: ideally, signs rendered and blended at video resolution before the video is rescaled; subtitles rendered and blended at display (or indeed any) resolution
• Clean up the karaoke tags?
• Easy text replacement syntax (no need for multiple tracks to en/disable honorifics)
• Bitmap graphics support
  • Use text as a mask over a bitmap texture
• Styles overhaul: Reduce to 2 fields: Style name and overrides, which are parsed as if in a {} at the start of each line with the style (e.g. Default,\fnCommie Sans)
  • Move Margins to an override tag; allow all 4 sides to have margins, instead of 3
  • Move Layer to an override tag
• Event overhaul: Remove unnecessary fields that have been moved to tags
  • Remove old Effects
  • Repurpose Effects field as general arbitrary metadata field (ignored by renderer; used by scripts and authoring tools as they like [Aegisub should decide standards for use?])
• Millisecond precision in timestamps (00:00:00.000)
• Move \move to \t-ing \pos
• Provide methods for specifying paths; keyframing
  • Avoid loads and loads of events for moving around a line
  • Animate text along a Drawing-style path?
• Don't be stupid about scaling
  • Same script works fine on 720p or 1080p files (and beyond!)
• \a tag (no 1-4) representing the alpha of all 3 bitmaps, applied after compositing
  • \fad will use this
  • This is almost certainly what you want anyway
• Put text on a path easily
• All arguments may be enclosed in (parens), rather than only most.
• Ruby character support (https://en.wikipedia.org/wiki/Ruby_characters)
• Require UTF-8 as the only valid character encoding
• Tags to enable and disable OpenType feature flags (ligatures, etc)

• Add tags for (at least) simple mathematical typesetting, i.e. superscript, subscript, fractions and matrices (only without brackets?). (original suggestion)
• Allow more flexible text justification; i.e. justify independent from event placement (currently possible only via libass API) and option for "full justification" for which both line starts and ends are aligned. (original suggestion)
• Movable clip position independent of the clip extent (#703)
• Add a way to create 3D-letters (original suggestion).
  However, this would be both costly to render and a lot of additional work to add, requiring at least:
  • letter thickness, i.e. extent in z-direction (as an addition: animatable scaling of thickness)
  • letter material (this can have quite a lot of variables; we can probably ignore anisotropic materials, but since letters can be partially transparent we probably can't ignore transmission behaviour)
  • lighting position(s) (& lighting colour?, only isotropic point lights or directed light and emitting surfaces...)
  • (probably better to have) its own event-type separate from regular (2D-surfaces in a 3D-space) dialogue and typesetting events.