Why This Is the Best Week Ever to Be an AI Buyer | Edition 275
Edition 0275 — Three labs, one Wednesday, and a price ladder that now runs from $1 to $50. Here's how to read the menu.
Something unusual happened last Wednesday: every major AI lab put its best publicly available model on the table at the same time.
On July 9, OpenAI opened its new GPT-5.6 family — Sol, Terra, and Luna — to everyone through ChatGPT, the API, and Codex. The same day, xAI made Grok 4.5 public. Anthropic had already made its move on July 1, when Claude Sonnet 5 became the default model for every Free and Pro user. Every major lab now has its best model out in the open, priced, and competing for your business.
But here's the thing: you're probably not going to comparison-shop across labs. You open Claude, or you open ChatGPT, and you stare at a dropdown with cryptic names. So the useful question isn't which lab — it's which of these do I actually click? Here's the menu inside each app, in plain English.
Inside Claude: a 4-model menu
Haiku 4.5 ($1/$5 per million tokens) is the fast, cheap one — quick questions, summaries, high-volume routine work. Sonnet 5 ($2/$10 through August 31, then $3/$15) is the default for a reason: it's now within a point of the premium models on independent tests, and it's the right choice for roughly 90% of daily work — writing, research, analysis, everyday coding. Opus 4.8 ($5/$25) is for hard, multi-step projects where the default visibly struggles. And Fable 5 ($10/$50) — the picker literally labels it “for your toughest challenges” — is exactly that: the model you reach for a handful of times a month, not the one you live in.
Inside ChatGPT: a 3-model menu
OpenAI's new lineup follows the same shape. Luna ($1/$6) is the volume tier — fast, cheap, fine for routine tasks. Sol ($5/$30) is the flagship: one point behind Fable 5 on Artificial Analysis's independent intelligence index, and #1 on their coding-agent index. And Terra ($2.50/$15), the comfortable-sounding middle option? Independent testing found it never wins on value — for any job, either Luna does it nearly as well for less, or Sol does it better for the same effective cost. Skip the middle.
| Model | When to click it | Input | Output |
|---|---|---|---|
| Claude Haiku 4.5 | Quick answers, high volume | $1.00 | $5.00 |
| Claude Sonnet 5 (intro to Aug 31) | Your default — most daily work | $2.00 | $10.00 |
| Claude Opus 4.8 | Hard multi-step projects | $5.00 | $25.00 |
| Claude Fable 5 | The toughest problems | $10.00 | $50.00 |
| GPT-5.6 Luna | Quick answers, high volume | $1.00 | $6.00 |
| GPT-5.6 Terra | The middle tier — tests say skip it | $2.50 | $15.00 |
| GPT-5.6 Sol | Hard problems + best at coding | $5.00 | $30.00 |
The rule that works in every app
Every lab now sells the same shape: a cheap tier, a default, and a premium. So the playbook is universal. Live in the default — it's the best value in both apps right now. Drop to the cheap tier for volume work. Step up to premium only when the default visibly fails you. And never assume the middle tier of a three-option menu earns its price — this week's independent tests just proved it doesn't. If your subscription is more than a few months old, this is the week to re-check it: the menu just repriced underneath you.