Gives five hill-climbing heuristics and released traces for tuning a coding agent harnessThe scaffolding around a model that turns it into a working agent — the loop, the tools it can call, and the rules for when to stop.Full definition → on open weightsA model whose trained parameters are published for anyone to download and run — unlike API-only models you can access but never possess.Full definition → models, where token overhead of 20-30% decides whether a cheaper model is actually cheaper.
Key takeaways · AI-distilled
Same model, different provider can change results: Cline's GLM-5.2 scored 74.2% on CoreWeave versus 61.8% via OpenRouter general routing, and they saw up to 2x differences in net tokenThe chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.Full definition → and cache hit rates between providers.
Identical reruns vary widely in Cline's runs, e.g. MiniMax M3 ranged 43.8% to 56.2% over five reruns, so the author reruns evalA repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.Full definition → and reports the distribution after ruling out infrastructure flakiness.
The author argues more reasoning effort is not automatically better, citing an Opus 5 FrontierCode result where medium thinking beat high, extra-high, and max, so the reasoning budget should be tested per model, provider, and task mix.
Terms in this piece · Glossary
open weights — A model whose trained parameters are published for anyone to download and run — unlike API-only models you can access but never possess.
agent harness — The scaffolding around a model that turns it into a working agent — the loop, the tools it can call, and the rules for when to stop.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
