Arabic has ~400M native speakers globally

Following up on this thread as an Arabic-speaking developer and long-time Claude user — wanted to add some context on why finishing this matters, and to say thank you for the progress already visible on the Code tab.
Scale: Arabic has ~400M native speakers globally. The Middle East AI market alone was valued at ~$15.6B in 2025 and is projected to reach ~$23B in 2026, growing at a 41.8% CAGR toward $265B by 2033 (Grand View Research). Government-backed AI adoption is accelerating fast — UAE’s National AI Strategy 2031 and Saudi Arabia’s SDAIA program are actively subsidizing enterprise AI procurement right now, and basic RTL usability is often a pass/fail criterion in those evaluations.
Why this is worth finishing, not just starting: Claude’s Arabic generation quality is already competitive — the remaining gap is purely at the rendering layer. That’s a rare case where the hard problem (model quality) is solved and what’s left is comparatively well-scoped UI work: proper dir/BiDi handling on message containers, mixed-language lines, tables, code/LaTeX isolation, and the input box.
Really glad to see native RTL land on the Code tab. Given the Chat tab implementation already appears to exist and is just gated, finishing that rollout — plus the remaining edge cases (per-cell table direction, mixed-language line detection, math/LaTeX isolation, and the input box/window layout on RTL systems) — would make Claude the first of the major labs to treat RTL as fully first-class rather than community-patched. Happy to help test on real Arabic/English mixed-content sessions if that’s useful.

Start by comparing the existing RTL implementation on the Code tab with the apparently gated Chat tab path. Inspect message containers, mixed-language lines, tables, code and LaTeX isolation, and the input box/window layout; done means RTL works across both tabs, including per-cell table direction and Arabic/English mixed content without disrupting code or math.

Escrita pelo modelo de indexação a partir do texto da issue.