Opus 5’s coding scores reportedly drop above “high” effort, not at max
hero88645 · reddit · 2026-07-25
Opus 5’s effort setting isn’t monotonic for coding
The thread argues that Opus 5 can get worse above “high” effort on coding tasks, not better. On FrontierCode, scores reportedly drop past the high setting because the model starts making unnecessary refactors and edits outside the requested scope.
It also cites two supporting data points:
- On AA-Omniscience, Opus 5 is about 11% more accurate than Opus 4.8, but its hallucination rate is roughly 6% higher.
- In CodeRabbit code review testing at xhigh, actionable-comment precision improved (39.3% vs. 35.2%), but recall on known issues fell (55.2% vs. 61.1%) and the model produced about 4× as many nitpicks.
The main takeaway is that the best effort level may be workload- and codebase-specific, and defaulting to max can waste money or even hurt output. The post also notes that when Claude’s safety classifier triggers, requests in Claude.ai / Claude Code / Cowork silently fall back to Opus 4.8 by default.
Related event: Opus 5 Coding Paradox: Higher Reasoning Leads to Lower Scores(13 posts)→
More from coding & agent
- 'My intelligence stack': developer splits AI coding work across Opus 5.5, Fable and Astra with human review gates — sh_reya · 2026-10-27
- Baseten webinar: evaluate open models for coding agents by benchmarking on your own merged PRs — baseten · 2026-10-27
- Thorsten Ball on working with agents: run more evals, automate everything, use the best models — omarsar0 · 2026-10-27
- Where to draw the approval line for scheduled agent jobs — Reddit discussion — MemStrata · 2026-10-27
- 4 MCP servers worth a look: full-text YouTube search, Chrome session reuse, Framer editing — Prior_Engineering486 · 2026-10-27
- DeepMind paper: LLMs can't sample randomly on their own — pass random numbers from code instead — rohanpaul_ai · 2026-10-27