PROMPT CHAMPS
Video guides / Claude Code

Claude Code / Prompt Champs

Which Claude Code Model? Fable 5 vs the Rest

04:49 runtimePublished July 5, 202610 key ideas

One developer handed Fable 5 an empty folder and asked for a working multiplayer game server - framework and all - and it built the whole thing in one sitting. That is the model now sitting at the top of your Claude…

Watch on YouTube

The complete breakdown

One developer handed Fable 5 an empty folder and asked for a working multiplayer game server - framework and all - and it built the whole thing in one sitting. That is the model now sitting at the top of your Claude Code picker. So you do what most people do. You pick the newest, biggest name and get back to work. Then the bill arrives, or a two-second typo fix takes the model ninety seconds to think about, and you wonder if you chose wrong. Here's the thing nobody tells you: these are not four sizes of the same model. They think differently, cost wildly different amounts, and answer on completely different clocks. One returns in a second. One will run for fifteen minutes on a single request and you are supposed to let it. Fable 5 is the most capable of the group, but capable is not the same as right for the task in front of you. Use it for a quick rename and you are paying ten times the price to wait ten times as long. So before you touch that picker again, here is exactly how Fable 5 differs from Opus, Sonnet, and Haiku, counting down to the one rule that tells you which to pick every single time. Number ten first.

Use this page with the video

Watch the episode above for the visual explanation, then use the notes below to revisit each idea, example, and practical move.

01

The context window

Fable 5, Opus 4.8, and Sonnet 5 each hold a million tokens - a whole repo plus its docs in one session. Haiku 4.5 stops at 200K. Handing Claude Code a large codebase at once? Skip Haiku; the other three keep the full picture.

Big repo in one shot: not Haiku.
02

What each one costs

Per million tokens: Haiku 4.5 near 1 dollar in, 5 out. Sonnet 5 about 3 and 15. Opus 4.8, 5 and 25. Fable 5, 10 and 50. The same task can cost 10 times more on Fable than Haiku. Match the model to the job, not to the biggest number.

Fable can cost 10x Haiku.
03

When you want it now

Renaming a variable, fixing a typo, one quick edit - you want the answer before you blink. Haiku 4.5 returns in a second or two while the bigger models pause to think. For fast, cheap, tightly scoped edits, keep Claude Code on Haiku.

Quick edits: Haiku, in seconds.
04

Fast mode is Opus only

Turn on fast mode and Opus 4.8 streams up to two and a half times more output per second - same model, less waiting - at a premium rate. Fable 5, Sonnet 5, and Haiku can't. Want speed without dropping to Haiku? That is Opus.

Fast mode: Opus 4.8 only.
05

Thinking you can toggle

On Opus 4.8 and Sonnet 5 you decide: think hard on a nasty bug, or switch thinking off for a snap reply. Fable 5 takes the switch away - it always thinks first. That buys depth, but you can't tell it to just answer quickly.

Fable always thinks - no off switch.
06

You won't see it think

Watching the other models reason out loud lets you catch a wrong turn early. Fable 5 never shows its raw chain of thought - you get a short summary, not the play-by-play. Trust it to run, but you steer with less visibility.

Fable shows a summary, not steps.
07

It runs for minutes

Give Fable 5 a hard task and one turn can run 10 to 15 minutes before it replies - planning, building, and checking its own work. Opus or Sonnet answer in seconds. Point Fable at overnight jobs, not a question you need answered now.

One Fable turn: 10 to 15 minutes.
08

Your daily driver

For the coding you do all day - features, refactors, tests - Sonnet 5 and Opus 4.8 balance speed, cost, and smarts. Turn the effort up and most work never needs a pricier model. Default here; escalate only when you get stuck.

Daily coding: Sonnet 5 or Opus 4.8.
09

The overnight refactor

A migration across 40 files, an hour-long build that checks its own work - that is where Fable 5 pulls ahead. Sonny Sangha had it one-shot a working Linear clone. Save it for the hardest job that finishes while you are away.

Fable 5: your hardest long job.
10

The rule to remember

Default to Opus 4.8. Drop to Haiku 4.5 for fast, cheap edits, and lean on Sonnet 5 for everyday builds. Escalate to Fable 5 only for your hardest, longest, run-it-alone tasks. Pick by the job in front of you, not the highest number.

Right model, right job, every time.

Video chapters

More practical AI guides, without the hour-long lecture.

Subscribe on YouTube