By skyforbes With GPT-5.1, the model reliably entered a deep reasoning mode when prompted carefully. In fact, I was able to use GPT-5.1 as a serious research assistant and recently published a paper in statistical physics applied to financial markets, where the model meaningfully helped with modeling intuition, derivations, and structure.
Since the rollout of GPT-5.2, I’m noticing a consistent change:
• Responses feel more generic by default
• The model often answers quickly with surface-level explanations
• Explicit prompts like “think deeply”, “take more time”, or “use extended reasoning” do not reliably route it into longer chains of thought
• There doesn’t seem to be a visible or controllable “thinking depth” option in the ChatGPT app (at least on Go)
My question is not about hidden chain-of-thought or internal reasoning disclosure. I fully understand why that’s abstracted away.
The question is about behavioral control:
How are users supposed to intentionally engage GPT-5.2 in longer, slower, research-grade reasoning?
Things I’ve already tried: • Longer prompts with explicit constraints
• Asking for derivations, assumptions, and limitations
• Framing the task as academic / research-oriented
• Iterative refinement
The model can still do deep work, but it feels less deterministic to trigger compared to GPT-5.1.
So I’m curious: • Is extended thinking now fully automatic and opaque?
• Are there prompt patterns that reliably activate it in GPT-5.2?
• Is this a product decision (latency, cost, UX), or just early-release tuning?
• Are Go users limited compared to other plans in how reasoning depth is routed?
I’m asking because for research users, the difference between “fast generic answer” and “slow structured reasoning” is massive.
Would really appreciate insights from others doing technical or academic work with GPT-5.2, or from anyone who understands how the routing works now.
Thanks.