Claude Opus 5.5 test: 18 of 18 critical defects fixed, 87.3 score
GoML benchmarked Claude Opus 5.5 on LatticeBench: the model fixed all 18 critical defects and 6 of 7 very_hard tasks, scoring 87.3. API pricing is $4 per 1M input and $20 per 1M output tokens, with a reported 40% cost reduction.
- Fixed 18 of 18 critical defects and 6 of 7 very_hard tasks
- AI Matic Bench score of 87.3/100, 7.5 points above the previous best
- API pricing: $4 per 1M input and $20 per 1M output tokens
- Modified 107 files, including settings outside the task scope
Read next
AI