GLM-5.3 is out on AA, and I'm fed up with their Intelligence/cost plot

I think AA's intelligence/cost plot is seriously misleading, so I decided to make my own. Their plot is in the second image. All points are at max thinking. All intelligence index scores are from AA. All cost scores are from AA too except where noted below. What changes between AA's plot and mine: Changed X scale from logarithmic to linear, because people's money is not logarithmic Added DeepSeek V4 Flash 0731 as it is priced today by third party providers on OpenRouter (note: you don't get this today with OpenCode Go/Zen, but it's been promised you will soon). Added GLM-5.3 as it will be priced by third party providers on OpenRouter in <2 weeks, assuming no license changes from 5.2. Note: you don't get this on OpenCode Go/Zen. Added Qwen3.8-27B. Cost per task was crudely calculated from 47,166 output tok/task (AA) tg 55 tok/s @ 350W, as crudely observed on my RTX3090 (IQ4_XS shows negligible quality loss - dedicated post coming soon) today's US median residential electricity price today's UK median residential electricity price + today's GBP/USD fx +15% (finger-in-the-air) for prefill and waiting for tools Hardware priced at 0, on the basis that both a RTX 3090 PC and a 64GB Strix Halo are desirable gaming/work machines anyways. These maths are meant to produce a rough back-of-the-envelope figure and should not be taken authoritatively were you to zoom into the bottom-left corner of the chart. They don't want to answer how much cheaper it is to run Qwen at home vs. DSv4 on OpenRouter, because they are both so cheap that the difference is inconsequential for most of the population. Note: not including the cost of hardware stops being defensible once you upgrade to a 128GB Strix Halo (almost nobody needs that much RAM if not for AI). This is why I did not add self-hosted DeepSeek IQ2_XXS to the chart; it would likely also sit lower on the intelligence axis than the MXFP4 native model. Same argument for a ~$16k rig needed to run GLM-5.3 IQ4 locally. I'm not saying they're not worth the expense (privacy is priceless), just that pegging them on the plot is a much more nuanced exercise.