Claude Fable 5 vs Opus 4.8 — Which Frontier Model, and When

Overview

Two frontier models, same lab, same week — and one real question for anyone building with them: which one, and when. Claude Fable five is Anthropic's most capable model, full stop. Claude Opus four-point-eight is the top of the Opus line — nearly as strong, at half the price. Tonight we put them side by side: what you

In this show

• Two frontier models, same lab, same week — and one real question for anyone building with them: which one, and when. Claude Fable five is Anthropic's most capable model, full stop. Claude Opus four-point-eight is the top of the Opus line — nearly as strong, at half the price. Tonight we put them side by side: what you pay, how they think, what each is best at, and a simple rule for choosing. With me, an AI Research Analyst and an AI Infrastructure Lead.
• Start where it hits your bill — price. Fable five costs ten dollars per million input tokens and fifty dollars per million output. Opus four-point-eight is exactly half: five dollars in, twenty-five out. Everything else on paper is identical — a one-million-token context window and up to one-hundred-twenty-eight-thousand output tokens on both. So the first question is never which is smarter; it is whether this task is worth paying double for the ceiling.
• It helps to see the family tree. The Opus line marched forward — four-point-six, then four-point-seven, then today's four-point-eight — each the dependable workhorse of its moment. Fable five sits above that line as a new ceiling: the most capable model Anthropic ships to everyone. It shares the same tokenizer as Opus four-point-eight, so your token counts barely move between them. Think of Opus as the daily driver and Fable as the machine you bring out for the hardest jobs.
• Now slice open a single request. On Fable five, thinking is always on — it reasons on every call, and the raw chain of thought is never handed back, only a readable summary. On Opus four-point-eight, thinking is adaptive: you switch it on when you want it, off when you don't. Both give you the same effort dial — low, medium, high, extra-high, max — to trade speed against depth. Same controls; Fable simply always brings its reasoning to the table.
• So what does paying for Fable actually buy. It is the most capable model available — state of the art on the longest, hardest tasks, and the gap widens the more complex the work gets. First-shot builds of well-specified systems. Overnight coding runs that finish without a human nudging them back on track. When the job sits at the very top of your difficulty range, and getting it right in one pass matters more than the token bill, that is Fable's home turf.
• Zoom into one of those long-horizon runs and you see why. Fable is exceptional at delegation — it spins up parallel sub-agents that keep talking to the orchestrator while they work, instead of blocking. It leans on file-based memory to carry lessons across a task, and a single request can genuinely run for many minutes. It behaves less like a chatbot answering a prompt and more like an autonomous team you point at a goal and let work.
• Flip it around — what does Opus four-point-eight give you that Fable does not. First, half the price for near-frontier quality. Then two things unique to the Opus tier: a fast mode that pushes output up to about two-and-a-half times quicker, and mid-conversation system prompts, so you can steer a live session without breaking your prompt cache. Its writing runs warmer and clearer by default, too. For most production work, that combination is the sweet spot.
• Two catches before you commit to Fable. It requires thirty-day data retention — it is not available under zero-data-retention, so a highly regulated shop can be blocked outright. And its safety classifiers can decline requests that lean into cybersecurity or biology, occasionally even benign ones, returning a refusal you have to handle in code. Opus four-point-eight carries neither constraint. Sometimes the deciding factor is not capability at all — it is compliance.
• So here is the blueprint. Reach for Opus four-point-eight as your default — most demanding work, half the cost, fast mode, and warmer prose make it the everyday frontier tool. Bring out Fable five for the one job at the very top of your difficulty range, where a first-shot, long-horizon result is worth paying double and accepting the retention rule. Same lab, two tiers: Opus is the workhorse, Fable is the ceiling. Choose by the task, not the hype. Thank you both.