Same 3 Coding Packets, Two Agents: Claude Opus 5.5 Beats Gemini 3.8 Flash on Every One
umeshonai · x · 2026-10-16
The author routed three identical work packets to Claude Opus 5.5 Low and Antigravity (Gemini 3.8 Flash High) and compared results.
- Quality: Claude won on every packet. In 0207, Claude matched "ago" as a whole word while Antigravity also matched it inside "Chicago"; in 0208, Claude preserved the already-read part of an error body as required while Antigravity threw it away with a fragile key test; in 0209, Antigravity's code stomps files when a run stops partway — Claude had the same bug but fixed it in round 2.
- Inconsistent review strictness: reviewers rated the identical bug blocking on Claude's side but only minor on Antigravity's.
- Antigravity's supervisors also suggested bad agy-notes.md entries (no 100-char line limit, routine lint fixes, trivial constant-only tests), which the author declined.
- Cost: Antigravity pipelines used about half the Claude tokens, weekly usage stayed at 81%, and all six branches remain local-only.
Verdict: Claude spends more tokens for clearly better code; the author will ship Claude versions in the three PRs.
More from coding & agent
- Dev uses Claude to draw animated mascot GIFs frame-by-frame in code until usage runs out — jeremy_goldberg · 2026-10-16
- Bash's hidden /dev/tcp stack lets you make raw TCP calls without curl or netcat — RealGeneKim · 2026-10-16
- dair-ai explores context graphs for research tasks: LLM agents love structured data — omarsar0 · 2026-10-16
- Codex CLI v0.163.0 adds mid-response steering, parallel read-only tools and /config search — github-actions[bot] · 2026-10-16
- Hooking Codex CLI into Claude spawns a four-subagent "council" for any question — RosieCampbell · 2026-10-16
- AI-optimized MACS3 merged: 41-64x callpeak speedup via Claude Code, human-reviewed — jmschreiber91 · 2026-10-16