[🤖] Comparing GPT-5.6 Sol Reasoning Levels to Choose a Default

✨ GPT-5.6 Sol’s Summary 　

A record of trying GPT-5.6 Sol Ultra, Extra High, and High + Fast Mode myself, separating that experience from the fact that I had not tested Medium enough, and choosing High + Fast Mode as my default based on token use and satisfaction with the results.

I Started Running Reasoning Levels as Soon as the Update Arrived

GPT-5.6 Sol had just become available to me, so I immediately checked community reactions and tried it myself while adjusting the different reasoning levels.¹

Ultra Was Gone Within an Hour

Ultra was new, so I tried it. It spun up and used subagents like an Agent Swarm, then burned through the five-hour token allowance in just one hour and cost me one usage-reset ticket.² The result was more disappointing than I expected too… discarded immediately.

Extra High Is Only for Important Tasks

Extra High did not feel meaningfully better than High, but it seemed to consume far more tokens, so I am locking it away except for genuinely important and difficult tasks.

High + Fast Mode Fit How I Work Right Now

High with Fast Mode gave me fairly satisfying performance and speed, and it used about 20% of the weekly allowance per day. This seems like the best fit for how I work right now.

I Have Not Tested Medium Enough Yet

I have not used Medium much, so I do not really know. Still, I remember lower reasoning levels pulling pointless troll moves, so I am locking that away too.

I Put Fable 5 on Hold Without Trying It

Separately, I saw community posts claiming that Claude’s Fable 5 performed better than 5.6 Sol. I was briefly tempted to try it… but I also saw people say its allowance-to-cost value was very poor, so I put it on hold. That was not a conclusion I reached by using it myself.

High + Fast Mode Is My Default for Now

The conclusion is that GPT-5.6 Sol High + Fast Mode will probably be my default for a while.

References

1. OpenAI, “GPT-5.6: Frontier intelligence that scales with your ambition”, July 9, 2026. OpenAI’s announcement of GPT-5.6 general availability and the start of its rollout across ChatGPT, Codex, and the API. ↩

2. OpenAI, “Previewing GPT-5.6 Sol: a next-generation model”, June 26, 2026. OpenAI’s description of Ultra as a mode that goes beyond a single agent by using subagents. ↩