The Latent NotesClear, illustrated notes on AI models, benchmarks and engineering.https://thelatentnotes.com/enThe cost of an AI task is more than its token pricehttps://thelatentnotes.com/articles/cost-of-an-ai-task/https://thelatentnotes.com/articles/cost-of-an-ai-task/A simple way to account for input, output, retries and tools when comparing models.Sun, 11 Oct 2026 00:00:00 GMTGemini 4 Argon: the numbers behind the launchhttps://thelatentnotes.com/articles/gemini-4-argon/https://thelatentnotes.com/articles/gemini-4-argon/A closer look at the output limit, token pricing and coding benchmarks — with the comparisons kept in context.Sun, 11 Oct 2026 00:00:00 GMTContext windows and output limits are different thingshttps://thelatentnotes.com/articles/output-tokens-and-context/https://thelatentnotes.com/articles/output-tokens-and-context/Two token limits, two different questions. Here is how to read a model specification without mixing them up.Sun, 11 Oct 2026 00:00:00 GMTHow to read a model benchmark without the hypehttps://thelatentnotes.com/articles/reading-model-benchmarks/https://thelatentnotes.com/articles/reading-model-benchmarks/A score becomes useful when you know the task, the evaluation setup and what counts as success.Sun, 11 Oct 2026 00:00:00 GMT