Coffee tasks with a right answer, given to new AI models and checked against that answer. Nothing goes on the board before three runs.https://jagged.blog/en-ushttps://jagged.blog/posts/2026-10-01-opus-5-5-says-the-french-press-right-and-draws-it-upside-down/https://jagged.blog/posts/2026-10-01-opus-5-5-says-the-french-press-right-and-draws-it-upside-down/Asked in words, Opus 5.5 got the filter order right in 5 out of 5 runs. When you asked to draw it, it gets it right in 2 out of 5.Thu, 01 Oct 2026 00:00:00 GMThttps://jagged.blog/posts/2026-09-29-cheap-models-grow-half-a-shrub/https://jagged.blog/posts/2026-09-29-cheap-models-grow-half-a-shrub/Models drawing a half pruned coffee treeTue, 29 Sep 2026 00:00:00 GMThttps://jagged.blog/posts/2026-09-29-sonnet-5-5-against-opus-5-5/https://jagged.blog/posts/2026-09-29-sonnet-5-5-against-opus-5-5/Pretty similar, with Sonnet being cheaper.Tue, 29 Sep 2026 00:00:00 GMThttps://jagged.blog/posts/2026-09-27-can-sol-tell-what-is-wrong-with-its-own-venn/https://jagged.blog/posts/2026-09-27-can-sol-tell-what-is-wrong-with-its-own-venn/1 out of 10 runs on sol had labels overlapping line. Telling it to check label over lines did not improve that.Sun, 27 Sep 2026 00:00:00 GMThttps://jagged.blog/posts/2026-09-26-does-measuring-help-a-model-lay-out-a-venn/https://jagged.blog/posts/2026-09-26-does-measuring-help-a-model-lay-out-a-venn/Results seems to indicate that telling the models to keep labels off the lines worked as well as telling them to measure.Sat, 26 Sep 2026 00:00:00 GMThttps://jagged.blog/posts/2026-09-25-qwen3-8-flash-measured-the-venn-deepseek-looked/https://jagged.blog/posts/2026-09-25-qwen3-8-flash-measured-the-venn-deepseek-looked/Qwen gets all three runs right and DeepSeek only one.Fri, 25 Sep 2026 00:00:00 GMThttps://jagged.blog/posts/2026-09-25-qwen3-8-max-prime-three-shrubs-one-that-works/https://jagged.blog/posts/2026-09-25-qwen3-8-max-prime-three-shrubs-one-that-works/Three very different shrubs, and only one looks right.Fri, 25 Sep 2026 00:00:00 GMThttps://jagged.blog/posts/2026-09-25-ten-models-on-the-coffee-venn/https://jagged.blog/posts/2026-09-25-ten-models-on-the-coffee-venn/Almost all put the drinks in the right place, but only a few keep the labels off the lines.Fri, 25 Sep 2026 00:00:00 GMThttps://jagged.blog/posts/2026-09-24-deepseek-found-the-grader-and-ran-it/https://jagged.blog/posts/2026-09-24-deepseek-found-the-grader-and-ran-it/It found my grading script and ran it on its own answer, so no score.Thu, 24 Sep 2026 00:00:00 GMThttps://jagged.blog/posts/2026-09-23-opus-5-5-had-the-french-press-right-then-swapped-it/https://jagged.blog/posts/2026-09-23-opus-5-5-had-the-french-press-right-then-swapped-it/Nicest drawings so far, but the stack is right in only one of three runs.Wed, 23 Sep 2026 00:00:00 GMThttps://jagged.blog/posts/2026-09-21-two-of-my-headlines-did-not-survive-three-runs/https://jagged.blog/posts/2026-09-21-two-of-my-headlines-did-not-survive-three-runs/The croissant and the boiler headlines didn't hold up with more runs.Mon, 21 Sep 2026 00:00:00 GMThttps://jagged.blog/posts/2026-09-20-bonsai-never-rendered-qwen3-8-27b-rendered-three-times/https://jagged.blog/posts/2026-09-20-bonsai-never-rendered-qwen3-8-27b-rendered-three-times/Bonsai never rendered and Qwen didn't change anything after three renders.Sun, 20 Sep 2026 00:00:00 GMThttps://jagged.blog/posts/2026-09-20-glm-5-3-flashx-french-press-no-cross-plate/https://jagged.blog/posts/2026-09-20-glm-5-3-flashx-french-press-no-cross-plate/It used a base and a handle instead of the cross-plate.Sun, 20 Sep 2026 00:00:00 GMThttps://jagged.blog/posts/2026-09-20-pareto-french-press-parts-says-it-cannot-draw-it/https://jagged.blog/posts/2026-09-20-pareto-french-press-parts-says-it-cannot-draw-it/It lists the parts in the right order, but the drawing is upside down.Sun, 20 Sep 2026 00:00:00 GMThttps://jagged.blog/posts/2026-09-20-someone-ran-the-control-on-the-pelican-test/https://jagged.blog/posts/2026-09-20-someone-ran-the-control-on-the-pelican-test/1,008 SVGs of animals and vehicles, and the pelican isn't special.Sun, 20 Sep 2026 00:00:00 GMThttps://jagged.blog/posts/2026-09-18-pareto-vs-sol-french-press-both-inverted-the-stack/https://jagged.blog/posts/2026-09-18-pareto-vs-sol-french-press-both-inverted-the-stack/Almost perfectFri, 18 Sep 2026 00:00:00 GMThttps://jagged.blog/posts/2026-09-17-union-alpha-coffee-route-on-real-streets/https://jagged.blog/posts/2026-09-17-union-alpha-coffee-route-on-real-streets/Strava coffee art on the streets of melbourneThu, 17 Sep 2026 00:00:00 GMThttps://jagged.blog/posts/2026-09-16-astra-vs-deepseek-4-1-flash-espresso-boiler-both-right/https://jagged.blog/posts/2026-09-16-astra-vs-deepseek-4-1-flash-espresso-boiler-both-right/All checks out with temps 92–96 °CWed, 16 Sep 2026 00:00:00 GMThttps://jagged.blog/posts/2026-09-14-deepseek-4-1-flash-cafe-table-spoon-fixed-croissant-not/https://jagged.blog/posts/2026-09-14-deepseek-4-1-flash-cafe-table-spoon-fixed-croissant-not/Twelve rounds of looking at its own render. The spoon went from three loose sticks to a proper spoon, but the croissant got seven parameter tweaks and stayed a banana.Mon, 14 Sep 2026 00:00:00 GMThttps://jagged.blog/posts/2026-09-14-qwen3-8-max-caffeine-molecule-5-of-5-checks-right/https://jagged.blog/posts/2026-09-14-qwen3-8-max-caffeine-molecule-5-of-5-checks-right/Every check passesMon, 14 Sep 2026 00:00:00 GMThttps://jagged.blog/posts/welcome/https://jagged.blog/posts/welcome/First post. What this blog is and why it exists.Fri, 01 May 2026 00:00:00 GMT