Just compared my Devil’s Advocate prompt across 5 different LLMs/Agents on a company I know well:

1. Claude 4.7 (regular)

2. ChatGPT 5.5 (regular)

3. Claude 4.7 (Research)

4. ChatGPT 5.5 (Deep Research)

5. Gemini 3.1 (Deep Research)

Conclusions:

• All were at least decent and usable

• Gemini was least useful (!), which is surprising because 6 months ago it was at the top

• The gap between regular and Research/Deep Research for both Claude and ChatGPT was small to medium - regular versions were pretty good

My rank order for this task was:

1. Claude 4.7 Research: 9/10

2. Claude 4.7: 8.5/10

3. ChatGPT 5.5 Deep Research: 8.5/10

4. ChatGPT 5.5: 8/10

5. Gemini 3.1 Deep Research: 7.5/10

Curious what you find if you are comparing workflows across different models/agents? Compound With AI - have you done any comparing recently?