Indie dev: Gemini4 Carbon matches 90-95% of Opus 5.5 for coding but is painfully slow
yihui_indie · x · 2026-11-16
Indie developer yihuiindie spent two days heavily using Gemini4 Carbon (burning 50% of an Ultra 20x quota) and compared it with Claude:
- Coding: ignoring Antigravity's poor harness experience, Gemini4 Carbon reaches 90–95% of Claude Opus 5.5's coding level
- Speed is the dealbreaker: with Claude he runs 4–8 parallel tasks; Gemini needs 8–12 parallel sessions to keep up
- Video: Claude Opus 5.5 and Fable 5.2 remain unmatched
- Copywriting: Gemini4 Carbon is the current SOTA at natural, human-sounding writing
The attached usage dashboard shows Codex weekly at 65% left, Claude session 30% / weekly 89% / Fable fully used, and Antigravity's Gemini at 63% (5-hour) / 49% (weekly). Verdict: decent output, but the slowness plus Antigravity's UX make it "gratuitous suffering" compared to Claude Code.
More from coding & agent
- How are people handling agent versioning, memory, and cross-device sync? — yangyi · 2026-11-16
- Vibe-coded movie seat picker hits 200K users in 24 hours after 3-hour build — itsOmSarraf_ · 2026-11-16
- Proof of Agent hits 1M onchain transactions and 10,000 agents in just 27 days — adamamcbride · 2026-11-16
- Running an always-on agent locally: 22 background runs used just 0.41% of quota across 1.51B tokens — Adorable-Cost-3249 · 2026-11-16
- Who's building agents that recover from mistakes? Partial failures across systems are the blind spot — djgoel123 · 2026-11-16
- ICL alone may enable substantial continual learning, plot shows agents improve by Month 3 — xennygrimmato_ · 2026-11-16