One vague prompt, seven upgrades, all eight prompts in full. Four add information; three change the process — and only one of those failures is easy to spot.

Google's own table puts it near the top. An independent leaderboard puts it 23rd for code. Both are true, and the builds show which one matters.

Codex is not a coding tool with a chat window bolted on. Almost every feature that separates it from ChatGPT falls out of one decision: it works on your disk, not on a copy.

Most model demos cannot be wrong in a way you can see. This one can — the numbers either match the source files or they do not.

Two products share the name, and the settings screen matters more than the build demo — because this one runs at 8am when nobody is watching.

A vault where the agent owns the wiki, may never touch the sources, and has to cite one for every claim it writes.

Ask the strongest model you have to write the operating orders for the cheaper one that replaces it — a succession letter, not a prompt.

Anthropic's pricing page makes a specific claim: Claude Opus 5 costs half what Claude Fable 5 costs. Not "a bit less" — exactly half, on both sides of the meter.

	Input / 1M tokens	Output / 1M tokens
Claude Fable 5 (claude-fable-5)	$10.00	$50.00
Claude Opus 5 (claude-opus-5)	$5.00	$25.00

You spend a week building a digital product. You polish the design. You list it. And then — almost no traffic.

The design was never the problem. The problem is that the niche was chosen by intuition, and intuition tends to land you in exactly the corner of the market where nine other people are already competing on price.