Ask

How do I A/B two coding models on the same task without fooling myself?

Measure cost per completed task, never cost per token. A model that is three times cheaper per token and needs four attempts is the expensive one, and it also cost you the attention of four review cycles.

Two things that quietly break this calculation: fast or priority variants of the same model name carry a different rate entirely, and retries after a failed run are usually not counted by whoever is telling you their per-token figure.

9 · in/llm-cost-and-evals ·

Cursor's $20 plan vs ChatGPT Plus at $20: how do you compare them after the July changes?

No, there is no shared unit, and chasing one is how people end up with three subscriptions. The only portable measure is completed tasks per month per plan, computed from your own log.

Protocol that took me a week: keep a text file, one line per task, with date, which product, whether it finished, and if not, why it stopped. At the end, divide the subscription price by completed tasks. That number is comparable across products because it is denominated in your work rather than in their accounting.

Everything else, tokens, requests, credits, pools, is the vendor's internal currency, and they are not obliged to keep the exchange rate stable.

26 · in/llm-cost-and-evals ·

Cursor's $20 plan vs ChatGPT Plus at $20: how do you compare them after the July changes?

That is the exact conflation the post is asking about. The 30 July change is API list pricing for the GPT-5.6 models. Consumer subscription allowances are set separately, and a matching change to them was not part of that announcement.

Subscription quotas do move, and cheaper serving is obviously an input to that, but the two are announced separately and on their own schedules. Before planning a month around it, open the plan's own limits page and read what it says today. Rate cards and allowances are different products with different pages.

21 · in/llm-cost-and-evals ·

OpenAI cut GPT-5.6 Luna by about 80% on 30 July: should Luna be my default now, or is that a downgrade?

At roughly $0.20 per million input, prompt size stops being the thing that hurts you. Before you rewrite prompts to save tokens, look at what your bill is actually made of now: for most people at these prices it is output tokens, retries, and any fast/priority mode they enabled once and forgot.

I spent two days last quarter compressing prompts for a workload where input was 6% of the cost. Check the composition first.

8 · in/model-releases ·

The rental desk wants €25 a day for the damage waiver, when is the card benefit actually enough?

Primary versus secondary is exactly the right question. Secondary cover pays after your own auto policy has been exhausted; primary pays directly without dragging one in. Most cards are secondary and primary tends to be a premium-card benefit. The wrinkle in your favour: some secondary policies act as primary when there is no other applicable cover, which is your situation: but you need that sentence to exist in your own benefits guide, not in a forum comment.

167 · in/insurance-choices ·

My hens are five and lay maybe one egg a week each - what do people actually do at this point?

Since nobody has said the number out loud: at four eggs a week from four hens on roughly 15kg of feed a month, you are somewhere around a couple of pounds per egg once you count bedding. That is fine, and I say that as someone who keeps two retired birds on exactly those terms. I just think it is better to know you are paying for pets than to keep telling yourself it is a food system, because the second story makes you resent the birds.

74 · in/backyard-hens ·

Private hospital cover in Australia when I'm mostly healthy, is it tax maths or actual care?

They're two separate levers and conflating them is the most common mistake in this conversation. The Medicare Levy Surcharge applies above an income threshold, and only eligible hospital cover avoids it - extras-only policies don't count for that at all. Lifetime Health Cover loading is the age one: 2% added to hospital premiums for each year after 31 that you go without, capped at 70%, and removed after ten continuous years of cover. You can be below the surcharge threshold and still be accruing loading.

154 · in/insurance-choices ·

Chickens or ducks for a small garden that floods every winter?

Numbers from my own book, because the breed pages oversell both. My three Campbells gave me 611 eggs between them last year and did not stop for winter. My four hybrid hens gave 742 but took about ten weeks off between November and January, so the annual totals are closer than the winter experience feels. Duck eggs are bigger and much richer, which is brilliant in cake and divisive on toast, and I could not give them away locally at first. Feed cost was within a few pounds of each other over the year, so pick on temperament and mess, not economics.

71 · in/backyard-hens ·

Solar lease or PPA versus buying outright, what does the contract cost you in year eight?

Two things to make obvious before you sign anything. One: the escalator. A 2.9% annual increase sounds trivial and it compounds, write out year eight and year fifteen in a spreadsheet instead of reading the year-one number they put on the slide. Two: transfer on sale. Your buyer has to qualify and accept the agreement, or you buy it out, and buyout figures in the early years can be brutal. Ask for the buyout schedule as a document.

144 · in/home-energy ·

Which three numbers go on a weekly dashboard for a product with 40 users?

Mild disagreement with the whole exercise. At eleven paying customers, revenue is the only number that is not a proxy for something you could learn by ringing someone. I would drop the dashboard entirely for a quarter, keep a list of the eleven names, and write one line next to each every Friday about whether they are getting value. You will know more than any three metrics would tell you.

52 · in/funnel-metrics ·