Carbon Matches Opus 5.5 for Coding but Slower, Says Dev — Though Frontend Design Still Sputters
iskander · x · 2026-11-16
Developer Ben Blaiszik shares hands-on results with the Carbon model: in the Antigravity harness it codes at roughly Opus 5.5 level, though slower, and he still slightly prefers Anthropic models for coding.
Key points:
- Before the Argon/Carbon generation, Gemini couldn't do any of the work he wanted — clear progress since.
- Frontend design base capability is still poor (defaults to "dark AI slop"), but given a design system it implements it flawlessly.
- He argues coding is becoming commoditized in late 2026, and is most intrigued by Carbon's research ideation, which feels top tier or on par with Fable 5.2 and Opus 5.5 — surprising him with plausible leaps where Opus and Fable get stuck.
More from coding & agent
- Matt Pocock's llm-wiki: ingesting tweets and community messages to decode his audience — mattpocockuk · 2026-11-16
- RELIC 0.8.0 shipped after years: maintainer says LLMs made the legacy crypto codebase manageable again — jedisct1 · 2026-11-16
- Berkopec: no one has built a good agent auto-memory, but LLM-wiki ideas may hold gold — nateberkopec · 2026-11-16
- A satirical year in AI coding: nobody writes code anymore, and we're still figuring out what that means — mattpocockuk · 2026-11-16
- Executor overhauls agent approvals with 4 modes, adds Gmail-style undo for any agent action — threepointone · 2026-11-16
- mcp-openrussian: a local read-only MCP server grounding LLMs in a real Russian dictionary — No_Television_5967 · 2026-11-16