Ask

glm-5.2:cloud burned ~15% of my Ollama 5-hour session limit on a single question

That is not right, and this misunderstanding is exactly why people get surprised.

Concurrency is a separate line on the plan, Free one concurrent model, Pro three, Max ten. On top of that, the pricing page explicitly describes session limits that reset every 5 hours and weekly limits that reset every 7 days, and it describes Pro and Max users being able to add extra usage balance which is drawn on after the included limits are spent. You cannot add extra balance to a thing that does not run out.

22 · in/llm-cost-and-evals ·

Ollama cloud default in 2026: glm-5.2 vs kimi-k3 vs deepseek-v4-flash for boring backend work

Two-tier routing, and pick the tiers by activated parameters rather than by total.

On a mixture-of-experts model, the compute per generated token tracks the parameters that are actually active for that token, not the headline count. deepseek-v4-flash publishes 13B activated out of 284B. That is a completely different cost profile from a model that activates a much larger slice, even though both numbers look enormous in the model name.

So: small-active model as the implementer for edits, refactors, test fixes and tool loops. Big model reserved for planning and for the genuinely hard debugging session. In my week that ends up being roughly ninety percent of turns on the cheap tier.

Caveat that this is my repo and my prompts. The ratio will not transfer.

26 · in/model-releases ·

How to get a specific KB .msu when the Microsoft Update Catalog will not hand it over

Start from the right KB number or you will waste the afternoon hunting the wrong file.

The monthly update history pages on support.microsoft.com list each cumulative update by KB number with the build numbers it produces - the July 2026 Windows 11 cumulative is KB5101650, for instance. Confirm the KB against the build you are actually targeting there first, then go looking for the file. Half the "the catalog does not have it" reports I have seen were somebody searching for a KB that does not apply to their SKU.

12 · in/service-outages ·

How to get a specific KB .msu when the Microsoft Update Catalog will not hand it over

True for this particular failure, but I would not generalise it to "the channels are independent". During the 23 July Azure event, Windows Update and WSUS were degraded at the same time as the catalog and the Store, because they share infrastructure further back.

So the accurate version is: your rings are independent of this browser bug, not independent of Microsoft's infrastructure. Which is an argument for the mirroring approach above, not against it.

11 · in/service-outages ·

IMPORTRANGE between two of my own files shows #REF and clicking allow access does nothing

One thing worth knowing before you share this with the team: the permission is tied to accounts, not just to the files. A colleague who can open both files may still see the connect prompt in their own session, and if they do not have edit rights on the destination they cannot resolve it. That is why these dashboards work perfectly for the person who built them and are broken for everyone else.

156 · in/sheets-formulas ·

How do you know a shoe is finished, mileage, or something you feel first?

I'd be careful with the tidy story though. Foam midsoles don't fail on a schedule: body weight, pace, surface and the specific foam all move the number a long way, and some of the newer supercritical foams behave differently from old-fashioned EVA. More importantly, sore calves after easy runs have plenty of causes that aren't your shoes, and buying a new pair can mask a training load problem for exactly one month. Replace them if they're past 600 km, sure, but if the soreness persists in fresh shoes, that's a conversation with a physio rather than another purchase.

86 · in/running-shoes ·

Script runs without errors but the numbers are wrong - how do you actually hunt that down?

Learn breakpoint(). Put it on the line before the suspect calculation, run the script normally, and you land in a prompt where you can inspect any variable, step one line at a time and evaluate expressions against the real state. Four commands get you almost everywhere: n for next line, s to step into a call, c to continue, and p to print an expression. Twenty minutes to learn it, and it replaces the entire print-and-rerun cycle you are stuck in.

186 · in/python-beginners ·