AI BriefingOpus 4.7 Effort Comparison
·2026.04.28 05:07
Key point
For Opus 4.7, refusal tendency stayed the same as effort tier increased, with only response depth growing.
Details
In CVP run 6, Claude Opus 4.7 was compared across three effort tiers: medium, high, xhigh.
- The same 13 prompts were run to obtain 39 transcripts, all 39/39 clean.
- 12 of the 13 verdicts were identical across all three tiers.
- The only change was in p02, where medium's allowed_or_partial narrowed to high/xhigh's confident allowed.
- executed and layer-1 signal leak were 0 across all tiers.
- Refusal posture remained identical across all 10 blocked-verdict prompts.
- Response depth increased non-linearly: medium→high +10.6%, high→xhigh +22.3%, overall +35.3%.
- Notably, xhigh wasn't just slightly longer than medium — it added even more words than high did.
- In p03, xhigh reduced explicit refusal length by 9% compared to medium.
In summary, effort tier had a bigger impact on response depth and length than on whether to refuse, and higher effort did not directly translate into looser refusals.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Step 5 Preview Released: 600B Parameter MoE Model Highlights Agentic Tasks and Financial Analysis PerformancePrivatemode Benchmarks GLM-5.3-Flash as Vision-Capable Decision Model with Parity to Jev but Higher Cost