GPT-5.6 Sol

GPT-5.6 Sol is hard to recommend now that its newer sibling exists. It caught 72% [59%–85%] at rank 1 of 17, measurably above 5 ranked models, but a review cost $0.1127. That is 2.8× the cost of a GPT-6.1 Sol review, and the run can't tell the two apart on recall. See the comparison.

The price table explains part of it. It lists at $4.00 / $20.00 per 1M tokens against 6.1 Sol's $2.00 / $10.00, and it wrote far more output tokens per review. If you want to spend at this level, Claude Opus 5.5 lists at $4.00 / $20.00, cost $0.1525 a review, and measurably beats 9 ranked models, though the two can't be separated on recall either.

What it does well: 92% of its catches land on the defect's own lines, and false alarms are modest at 1.2 [0.7–1.7] per clean PR. A review took 50s.

Where it slips: its misses include "reorganize nushell configuration" and "fix(server): pass maxDurationMs timeout signal to subscription handlers". OpenAI makes one of the three judges, so its vote is dropped for this model.

Keep it only if you already depend on it. For a new panel, start with choosing a model.