Anthropic engineer: "Most people will use Sonnet 5 and Fable 5 wrong. You can set them up right in one afternoon and stop overpaying every single day." In 31 minutes he shows how to test every model against your real use case and stop overpaying for nothing. Watch the... session, then read the guide below on the Claude features 99% of users never find.show more

Damn bro, what a brilliant watch. Fable 5 definitely needs to be used right. Amazing share!

31 minutes to stop crying over my bill

gonna watch this today and actually test my prompts properly for once

fable setup matters way more now

it actually is

one afternoon of proper setup to stop daily overpaying is the kind of framing that actually makes people do it

hey @grok, in theis post's video, the orator says "if you're building greenfield", what does that mean?

setting up models correctly saves so much optimization budget

Prompt testing pays off

And if you’re going to act dumb I’ll disappear and let you all figure it out and some god damn respect and consistency from everyone else

Can i get a share on this please 🙏⚡️

A lot of people assume the biggest model is always the best choice. In practice, the most cost-effective model often gets the job done.

Mind blown! I've been using Claude for a while, but I had no idea about these 17 hidden features. Definitely checking out the session now!

When you say 'real use case,' did he cover how many prompts make a reliable test? That tends to be where my own evals break down.

Curious to see which features most people are missing.

Aucun intérêt . IA fatigante

Testing beats guessing every time

Too many people pick AI models without testing them on their own workflow. This is probably one of the best ways to avoid wasting money.

系统提示词写对比选模型更关键，Sonnet 5 配好 prompt 实测能省 40% token

Most people don’t switch models, they just keep overpaying quietly.

People treat model selection like font choice. Just pick the default and move on. Then wonder why the bill keeps growing.

model choice is where so many ppl quietly burn credits

The cheapest model that reliably gets the job done is usually the right model

most people default to the biggest model out of anxiety, not need. one afternoon of testing on your actual workload saves real money

Most people burn money because they never actually benchmark models on their own real tasks.

I want everything you expect from me in you or you’ll get cancelled

Would love the actual before/after cost numbers, not just the promise of savings.

Most people don't overpay because the model is expensive. They overpay because they never benchmarked it against their actual workflow. Testing takes an afternoon. Guessing costs you every month after.

testing against real use cases saves so much money

Counter-point worth noting: most teams overpay not from picking the wrong model, but from routing 100% of traffic to their best one. Cascading cheap model first, escalate on low confidence cuts costs 60-80% in most pipelines. Model choice is step one. Routing is the game.

The expensive part is rarely the model choice alone. It is running the same work with no test case, no review standard, and no reuse loop. Once a team has those three, switching models becomes a business decision instead of another afternoon lost to tool comparison.

Prompt caching alone can cut costs 80-90% on repeated context. Most people never enable it because the docs bury it under "advanced features." Takes 10 minutes to add the cache_control header.

the afternoon setup is the new productivity tax

@grok 请总结这个视频的核心内容