Opus 4.8 can run in a Fast mode that's about 2.5x quicker and roughly 3x cheaper than earlier fast modes — or in full deep-reasoning mode. The skill is matching the mode to the task.
Opus 4.8 isn't one fixed speed. Fast mode trades a little depth for a lot of speed — around 2.5x quicker — and it's roughly 3x cheaper than the fast modes of previous models. Deep mode gives you full reasoning for the hard stuff.
Same model, two gears. Most people leave it in one gear and either overpay or wait too long.
Fast mode is built for high-volume, low-ambiguity work: quick edits, formatting, simple lookups, summarizing a thread, cleaning up files. When the task is clear and you're doing a lot of it, speed and cost win.
Paying for deep reasoning to rename files is like hiring a surgeon to cut your sandwich.
fast: reformat 50 notes fast: summarize a thread fast: rename + sort files
Deep mode earns its keep on tricky reasoning: system design, gnarly debugging, strategy, long-horizon multi-step problems. When being right matters more than being quick, give the model room to think.
Slow is smart exactly when the stakes are real.
A simple habit: start fast, and escalate to deep only if the task stalls or turns out to be harder than it looked. Reserve deep for the ~20% of work that's genuinely hard, and let fast handle the rest. You'll save real time and money without losing quality where it counts.
The whole skill is knowing which gear you're in.
A one-glance guide to picking Fast or Deep mode, with examples and an escalation rule.