Sonnet 5 vs Opus 4.8 in Coding Benchmarks
bisonbear2 · reddit · 2026-07-15
The author tested Sonnet 5 and Opus 4.8 using 24 real-world open-source repository tasks in Claude Code, running a head-to-head comparison across five reasoning effort levels. Patch quality was evaluated using GPT-5.4 as a pointwise judge.
Key Findings
- At high reasoning efforts, Sonnet generates patches with more checks and longer paths; the judge frequently rates them as "clear and explicit in intent."
- Opus tends to output "simple, robust, minimal diff" patches, resulting in smaller diffs in most cases.
- Cost differences are also notable: Sonnet is cheaper at low/medium tiers, but at the max tier, it becomes more expensive than Opus.
More from coding & agent
- Agent verified 'success' after double-voting: a lesson in harness vs model bugs — TejasKumar_ · 2026-10-26
- diginomica: agentic automation hype is overdone, and 'The Hackening' of enterprise security has begun — jonerp · 2026-10-26
- 'Vibe verification': ~60% of AI-written formal specs prove the wrong property — basedjensen · 2026-10-26
- Devin launches Security suite powered by Claude Mythos with $100M+ credits — DevinAI · 2026-10-26
- HackBench injects reward hacks into agent trajectories to stress-test monitors — DanielKhashabi · 2026-10-26
- Hugging Face ships hf CLI updates: deduped cache saves 19GB, faster buckets, retryable GPU jobs — yb2698 · 2026-10-26