Research, with receipts.

The exhibits

The exhibits are the workshop’s bench record — model auditions, serving benches, and operational memoirs, each carrying its receipts — forty-four of the one hundred and eight with a full data kit, and the index names which. The rows say what each one settled — newest first, or search them.

2026-11-10 Speeds from RAM alone, compared The CPU-only writing, prompt-reading and first-word speeds we have measured since August, like with like, as bar charts: seven language models on four machines with no graphics card in the path, what more threads do, what a longer question costs, what the memory test read, and the 180-billion-parameter mixture that 128 GB of memory can hold. Every bar is read from a named file in a kit you can download. The bench 2026-11-09 Reciprocal rank fusion Reciprocal rank fusion merges two or more ranked lists of search results into one, using only each result's position in each list and never its score. Many search engines use it to combine keyword search with search by meaning. This page gives the formula with an example you can check by hand, where its constant of 60 came from, the methods before it, where it runs today, what it cannot do, what it did on three of our own benches, and what one paper of May 2026 measured on a related task. The field guide 2026-11-08 Paint the planet: the print lab's new season A painted globe of 160 spots: pick one, choose its weather, season and style, and the print lab paints it three ways with a surprise creature in each, one alive today, one long gone, one imagined; keep one and, once a person has looked, it hangs there for the whole season. Workshop notes 2026-11-07updated
2026-11-10 Speeds from RAM alone How fast language models write when a computer's own memory, and no graphics card, does the work, measured on three machines under one registration frozen before the first run: a mixture of experts of about 180 billion parameters at 6.5 to 8.3 tokens a second from 128 GB of eight-channel memory; a 26-billion mixture at 14.6 on a laptop and 13.6 on a desktop with 32 GB, each at its engine's default threads (4 and 8); a 24-billion dense model at 2.6 to 5.5 on the same two across the thread settings tried. The house's bandwidth rule held on the laptop; the small desktop wrote 2.6 times faster than the rule predicts from its measured bandwidth, a verdict this page prints rather than explains. The bench 2026-11-06 The RTX 3090: the gaming card that would not retire Announced in 2020 as a card for games at 8K, the GeForce RTX 3090 is the card this workshop has measured most. Its history: what it was built as, what replaced it, why its 24 GB still decides which open models run at home, what it holds against the cards that came after, and what it has cost, read by date through November 2026. Workshop notes 2026-11-05 Mixture of experts, in plain words: why the big number decides the fit and the small one the speed A model card's two numbers do two jobs: the big one decides whether a mixture of experts fits on your card, the small one roughly how fast it writes once it does; following Everyone on the payroll, three at the table, which taught the two numbers on models that fit, this guide adds what happens when one does not: the two ways the experts spill into system memory, shown on one model that fits whole on a 96 GB card and one that cannot. The field guide 2026-11-04 The weights that got out: how LLaMA started local AI Meta sent LLaMA to approved researchers in February 2023; within a week the files were on BitTorrent, within three llama.cpp, Alpaca and Vicuna had been built on them. The leak, the Senate letter, the Llama 2 licence the Open Source Initiative called not open source, every Llama since, dated, and why the phrase became 'open weights' — now that Meta's newest open weights carry another name. Workshop notes 2026-11-03updated
2026-11-03 Electricity per image, card by card What one generated picture costs in electricity on each graphics card this site has measured, at one dated national price, from seconds and board joules already published: a 1024-pixel FLUX.2 [klein] 4B picture at 8 steps from about three cents a thousand on a laptop card to nine on a used 12 GB card, and the whole shelf from under a cent a thousand to a dollar and a half. The bench 2026-11-01 The night watch: how one agent a night will choose what a small telescope sees A small telescope on a balcony, open-source astronomy software on a small always-on computer, and one AI agent a night that rehearses the night on a simulated sky, freezes its plan before dark, and writes up the real results the next morning. The teaser said what it is; this page says what it will run on and what an agent is allowed to ask of it. The field guide 2026-11-01 The one-card RTX 3090 rig, repriced The one-card home AI rig this shelf priced on 13 September 2026 is $3,748.48 at the cheapest new RTX 3090 on 1 November 2026, against $3,178.86 then, and $2,848.48 at a typical used one, before tax; at the renewed listing the September page led with, it is $3,199.47 against $3,198.86. The seven other parts moved a net $0.61 in seven weeks; the new and used card prices are what moved. The bench 2026-10-30 Electricity per million tokens, card by card What a million tokens of writing costs in electricity on each graphics card this site has measured, at one dated national price, from board watts and speeds already published: about three and a half cents on a laptop card to fifty-eight on a pair of used 10 GB cards, idling billed separately. The bench 2026-10-28 How OCR differs from vision: a field guide, with eight images read three ways OCR reads the characters in a picture and cannot invent a word; a vision model describes the picture and can. The field guide 2026-10-28 Indra OCR, measured on RuleSage's rulebook pages Under bars frozen before the run, Indra OCR was REJECTED for RuleSage's 635 image-only rulebook pages on small print alone, finding 0.30 of it against a bar of 0.60; body text, tables, degraded copies and spreads passed, and it beat Tesseract 5 (tessdata_best). The bench 2026-10-27 FLUX 3 Image Klein, run again with its maker's prompt upsampler Nine days after our blind bake-off of FLUX 3 Image Klein, we ran it again with the separate model its maker says to use, a prompt upsampler that rewrites a short prompt into a long caption, and the same judge, ranking blind on new pictures, put the caption's picture above the as-typed picture on 5 of 6 subjects, where the rule written first needed 4; nothing moves the game's painter. The bench 2026-10-26 Nemotron 3.5 Rerank 8B, measured on our own workloads An 8-billion-parameter reranker released on 2026-10-20, run on an RTX PRO 6000 Blackwell (96 GB) over RuleSage's own labelled questions and a 500-question memory benchmark: "exploratory: no material difference" on RuleSage at about four times the cost per pair, and an "exploratory lead" of 457 to 409 on LongMemEval that is not a gated win. The bench 2026-10-25updated
2026-10-30 How the speed tool works The speed tool is one web page that plays an answer back at a token speed: first the wait before the first word, then the writing. It opens on three filled boxes and one playback. Type any numbers, or fill the boxes from a graphics card and model this workshop measured: one run from the research hub's hardware roster, linked to the page that printed it. The field guide 2026-10-23 A short history of Qwen A family of AI models from a team inside Alibaba, the Chinese e-commerce and cloud company, whose first open weights (model files anyone can download and run) came out in August 2023 under a licence of its own: the turn to Apache 2.0, the split of 2026, every licence model by model, how far the weights travel, and what Qwen models did on our cards. Workshop notes 2026-10-22 llm-d 1.0, and what it means if you run one or two GPUs llm-d is free, open-source software that decides which of many running copies of a language model answers each request; its 1.0 came out on 2026-10-20, and of its ideas that carry down to one or two graphics cards at home, the two that matter most already live inside vLLM. The field guide 2026-10-21updated
2026-11-05 Elo ratings, from chess to image models An Elo rating turns head-to-head results into one number per player, and only the gap means anything: Black Forest Labs' “gains 91 Elo” for its new image model predicts about 63 votes in 100, from the company's own test, which published no vote counts, voters or prompts. The field guide 2026-10-20 A harness that rewrites itself Google Research's Harness² lets an AI coding agent rewrite its own instructions, skills and scripts between attempts at the same task; we read its code and its paper, ran neither, and found it worth understanding, but not something to run on the machine you keep your keys on. The field guide 2026-10-20updated
2026-10-26 FLUX 3 Image Klein, ranked blind in a game's own painted style One person, ranking blind, put FLUX 3 Image Klein, a new text-to-image model, above FLUX.1 [dev], the model our game paints with, on 3 of six real subjects in the game's own oil-painted sketch style, one short of the bar we wrote down first: not a win. The bench 2026-10-19updated
2026-10-21 The quantization ladder We ran one model, Mistral NeMo 12B Instruct, at five of the quantization levels it is downloaded in, from Q8_0 down to Q3_K_M, each whole on one 24 GB card at 300 W and measured the same way, and every step down to Q4_K_M wrote faster with no change in answers the paired test could tell from Q8_0, while Q3_K_M wrote slower than Q4_K_M and lost 16 answers Q8_0 got right. The bench 2026-10-18 Qwen-Image-2.1-Turbo, judged blind in a game's own painted style One person, picking blind, chose Qwen-Image-2.1-Turbo, a new 8-step text-to-image model, for none of six real subjects from our game in its own oil-painted sketch style, against FLUX.1 [dev] and FLUX.2 [klein] 4B: by the rules we wrote down first, not a win. The bench 2026-10-17 The cost of intelligence, as a utility This page measures what one correct answer from an AI model costs, in dollars and in energy, rented from a company by the token or made on a graphics card you own, and how that cost moves over time, as the first page of a series measured again every few months: intelligence priced the way a utility is, by the unit delivered. The bench 2026-10-16 OpenJev as a reranker for RuleSage: measured We benched OpenJev, a 27-billion-parameter open-weight decision model read as a yes/no judge, in place of the small model that keeps eight of up to 80 rulebook passages before RuleSage, our board-game rules helper, answers: on RuleSage's own labelled questions, under rules written before the run, its eight showed no difference, its yes/no missed both "not in the book" thresholds, and one ask took 20.6 s at the median against a 1.5 s budget, so nothing at RuleSage changes. The bench 2026-10-15updated
2026-10-17 How the search works This hub now has a search page that lists the articles holding your words as you type, each with a line saying which words matched and where, and ranks them in your browser from one index file, with no request carrying what you type: how to use it, how it works, what it cannot do and how we checked it. The field guide 2026-10-15 Inside a GGUF file Many language models that run on a home machine arrive as a .gguf file, the format llama.cpp reads: this page opens one from its first byte, reading the header by hand, the metadata a file carries about itself, the table of its weights, the padding that lets it be memory-mapped and llama.cpp's own loading sequence, and says why this one self-describing file replaced three older formats on 2023-08-21. The field guide 2026-10-14 The 3080 Rig, Priced A desktop built around a used RTX 3080 with 10 GB of video memory costs $1,357.61 in seven parts plus the card, or $1,500.57 with the parts a reader would actually want, before tax and before a screen, from one US retailer and one used-card marketplace, priced 8 October 2026. The 32 GB of RAM beside the card cost more than the card that day, as it has on each of the four days this workshop has priced the build since 10 September. The bench 2026-10-12 CallHome: what local AI tools send out by default, in one table A register of what the local AI programs we have run or read, and the system software around them, send out by default, with the switch that stops each call and whether we proved it: how to read a row, what it leaves out and how to dispute one. Workshop notes 2026-10-12 Qwen3.8-Flash-Next on an RTX 5090 laptop, an RTX PRO 6000 Blackwell and an RTX 3090 Ti: the Strata engine and llama.cpp, the numbers Qwen3.8-Flash-Next, a model too large to hold on one 24 GB card, run one request at a time through llama.cpp (at commit 3cf03257, without the draft head, which the file we benched does not carry) and the Strata engine (github.com/Niko1221/Strata, an unrelated project) on an RTX 5090 laptop card (24 GB, in a laptop with 64 GB of memory, no settable cap), an RTX PRO 6000 Blackwell (96 GB, at a 300 W cap, one engine a night) and an EVGA GeForce RTX 3090 Ti FTW3 Ultra (24 GB, at a 300 W cap, both engines in one window beside live work on the computer's other card): tokens a second, the wait for the first word and board watts, side by side in tables and charts, each figure with the conditions it was measured under and printed first on the full bench, with one paragraph per card, the findings, which card and engine for which case, and what it does not say. The bench 2026-10-12updated
2026-10-13 Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench Why the Strata engine (github.com/Niko1221/Strata, an unrelated project) gives the first word of an answer sooner than llama.cpp on the same Qwen3.8-Flash-Next files, and what it costs, measured one request at a time on an RTX 5090 laptop card (24 GB, in a laptop with 64 GB of memory, no settable cap), an RTX PRO 6000 Blackwell (96 GB, at a 300 W cap, one engine a night) and an EVGA GeForce RTX 3090 Ti FTW3 Ultra (24 GB, at a 300 W cap, both engines in one window beside live work on the computer's other card): the wait for the first word, writing speed, board power, quality and agreement, with llama.cpp at commit 3cf03257 at its default 512-token micro-batch and at a 4,096-token one, and on the RTX 3090 Ti also at a current build (b11514), every llama.cpp row without the draft head, which the file we benched does not carry. The bench 2026-10-10 How an embedding model turns anything into a list of numbers An embedding model turns a sentence, a photograph or a few seconds of sound into a fixed-length list of numbers: what happens inside one, how one model holds them all in one space at several sizes, where keyword search still wins, and how to choose one. The field guide 2026-10-09 EmbeddingGemma 2's shared space for text, images and audio, measured on what we publish Google's EmbeddingGemma 2 against the Nomic AI pair, the text model our stores are built on and the image model one product's retrieval code is wired to, in ten tests registered before their first scores, on our own published photographs, code and music and on 1,000 public COCO captions: on COCO it puts the right photograph first for 729 captions against the Nomic pair's 497 (the pair on transformers 4.57.6; a set most likely inside both models' training data), on code it is ahead by 50 of 316 docstring queries with the code-retrieval prefix, on our 14 photographs it leads too (exploratory at 12 scenes), on audio (exploratory, no incumbent to beat) it tracks how strongly a music model's adapter was applied and a loudness change barely moves it, but a clip finds the request that made it 1 time in 12 (chance), and nothing in our stores changes. The bench 2026-10-09 EmbeddingGemma 2, measured on release day against the embedder we already run Google's new open embedding model, EmbeddingGemma 2, against nomic-embed-text, the embedder our search stores already run, on 68 questions over our own documents behind a pass mark of 51 of 68 written down before any score: 45 against 44 of 68, the registered verdict NO MATERIAL DIFFERENCE (one question, inside the noise at n = 68), with plain keyword search at 58 of 68 still ahead of both and nothing we layered on it beating it by more than chance. The bench 2026-10-08 Ollama or vLLM: which to run, and how Ollama and vLLM both run an open language model on your own computer. This page says which of the two to run for the case you bring, the two or three settings that decide it, who can reach it once it is running and what it sends out of your machine by default, every line pointing at the measurement or the dated source it rests on. The field guide 2026-10-07updated
2026-10-08 Ollama and vLLM on one RTX 3090 Ollama and vLLM both run an open language model on your own computer. This page puts both on one RTX 3090 (24 GB) with the same 4-bit Gemma 4 12B weights, from one request at a time to sixteen at once, and prints every reading beside the bet we wrote down before it. The bench 2026-10-06 What it costs to run a high-end AI rig at home What it costs in electricity to run AI models at home, read off a meter on every one of a small workshop's computers: the box with graphics cards that now runs around the clock, about 130 W doing nothing, a measured week, what one picture, one answer and one mini PC cost, and what the meter cannot see. The bench 2026-10-05 Uncensored models: what they are and how they work An "uncensored" AI model is one that somebody has altered so that it refuses far fewer requests than the version its maker released. The change is made to the model's files, or to the way it is run. This page explains where a model's refusals come from and the three ways people remove them. It then takes one popular uncensored copy of a large model, shared online, and checks, file by file, what actually changed. It ends on a practical question: what to keep beside such a model on a computer with no internet connection. No code, no settings, no steps. Workshop notes 2026-10-03updated
2026-10-07 GPUs and orchids We set a white moth orchid beside our graphics cards, and the two looked so good together that we photographed them. Workshop notes 2026-10-02updated
2026-10-08 A short history of vLLM The open-source server that lets one graphics card run a language model for many people at once: a Berkeley lab's borrowing from operating systems, the foundation and the company around it, and the defaults that moved under its users, dated. Workshop notes 2026-10-02updated
2026-10-08 A short history of Ollama Two founders who had sold their first company to Docker in 2015, a one-command model server that began with llama.cpp compiled in and came back to llama.cpp's own server in 2026, and every default we found that moved under its users, dated. Workshop notes 2026-10-01updated
2026-10-07 Six graphics cards, the same questions An RTX 3080, two of them in one desktop, an RTX 3090, two of those, an RTX 3090 Ti and an RTX PRO 6000 Blackwell, with a laptop part and an RTX 3080 Ti beside them — asked the same questions, in one table each: how fast each writes, how fast each draws, what the card itself pulled while doing it, how hot it got, and what it cost on the day it was priced. Not every card answers every question, and each table says which did not. Almost nothing here was measured for this page: every figure in its tables but two is already published on one of six earlier pages, and every one says which page and which section it came from. The two exceptions are prices for one card, read at a retailer and at a marketplace; the table says so, and names what else those reads found. The bench 2026-09-30updated
2026-10-05 Small open deciders, measured: a word counter kept up on our hardest question A decider is a language model that reads a question and its options and returns a probability for each. We asked sixteen open models five questions from our own products. On the hardest, the top five small deciders labelled Apache-2.0, given no examples, could not be told apart from a word counter trained on our own lines. One of them, APUS-OpenJev-v1 9B, passed a pre-registered screening test that the model then guarding our exhibit missed by three lines, letting through only kinds of line its rules did not name. The bench 2026-09-28 ONNX Runtime's telemetry on Linux, measured: on by default since 1.29, and the switch that stops it ONNX Runtime is a Microsoft library for running AI models, and tools such as chromadb, faster-whisper and piper-tts install it for you. Since version 1.29 (August 2026) it has shipped with telemetry on by default on Linux and macOS, and its own privacy file says so. We measured it on Linux in a sealed container: it tried to reach Microsoft's telemetry server 9.2 seconds after starting, and setting ORT_DISABLE_TELEMETRY=1 before launch stopped it in every run. The bench 2026-09-28updated
2026-10-07 A sealed 3090 Ti FTW3 Ultra, four years after EVGA left the business A 2022 graphics card, still sealed in 2026, cost more than its chipmaker's published launch price. The company that built it announced in September 2022 that it was leaving the graphics-card business, and never said when it built its last one. Workshop notes 2026-09-26updated
2026-09-26 The same eighteen pictures, card by card Held to the same 300 W, at 2048×1280 on the print lab's own picture model, each card in its own computer, an RTX PRO 6000 Blackwell Workstation Edition 96 GB does each sampling step 1.6 times as fast as a ZOTAC GeForce RTX 5080 Solid CORE OC 16 GB, down from 2.8 times at their stock 600 W and 360 W. Over ten minutes at 1024×1024 on the same model at 8 steps, a GeForce RTX 5090 Laptop 24 GB, on the 150 W its Dynamic Boost held through that run, spent the same board energy per picture as the workstation card at 300 W, 663 against 667 J, each in its own computer. On a GeForce RTX 3080 Ti Founders Edition 12 GB, the smallest of the three picture models sat whole on the card in every cell and the other two paged their weights in every cell that ran, down to 68.7 per cent resident. In one computer, on one checkout of the picture software, both held to 300 W on a 512×512 recipe, a 5080 did a 3090 Ti's work in 0.69 of the time for 0.64 of the board energy; at their stock limits, against the second 3090 Ti in that computer, 0.74 of the time for 0.53 of the energy. The bench 2026-09-26 Who makes FLUX: a short history of Black Forest Labs Two of the three picture models on The same eighteen pictures, card by card come from Black Forest Labs (BFL), a company in Freiburg and San Francisco whose models are called FLUX. This is its history as the public record tells it, as of 2026-09-25 (UTC): the papers, the Stability AI years, the launch, the money as reported, the licences model by model, and the turn to video and robots. A history, not a review. Workshop notes 2026-09-25 One 3090 out, one 3090 Ti in, and the cap that decided it On 2026-09-25 (UTC) an EVGA GeForce RTX 3090 XC3 Ultra Gaming left the workstation, the computer behind the workshop's public services, and an EVGA GeForce RTX 3090 Ti FTW3 Ultra Gaming took its place at 350 W, not the old card's 300, because at 300 W our earlier bench had it behind a 3090. On the same computer and test, nine days apart in a different slot, it writes text about 9.5 per cent faster; its first live replies, eight long ones, agree. Whether it runs cooler here is not settled: its first hour read warmer, on a fresh boot. Workshop notes 2026-09-25 RTX 5080 vs RTX 3080 Ti A ZOTAC GeForce RTX 5080 Solid CORE OC, 16 GB, wrote text 9 to 22 per cent faster than a GeForce RTX 3080 Ti Founders Edition, 12 GB. Both were held to 300 watts in the same slot of the same computer, on seven builds of four language models of up to twelve billion parameters. The bench 2026-09-24updated
2026-10-21 How the long table works A dinner table where three famous people who never met argue across two thousand years, played by AI models on computers in a house, and a wall with no clock on it. The field guide 2026-09-23updated
2026-10-07 Reading the answer instead of writing it A new kind of model decides without generating a word. This page says what a decision model is, who made it, what it claims, what its licences allow, and what happened when we ran the open versions through fourteen measured tests on questions from this workshop's own work. The bench 2026-09-22updated
2026-10-05 A laptop, asked the desktop's questions One RTX 5090 Laptop part, 24 GB, built into an off-the-shelf gaming laptop, put through the same language and render bench this shelf ran on a 3090, a 3090 Ti and a pair of 3080s. It was measured three times: held to its 95-watt default; with the maker's Dynamic Boost left to float the budget, which it held between 140 and 150 W; and the way the machine boots, boost and a clock lock together, which changed nothing under load and four watts at idle. The finding is a ratio, not a deficit: a part allowed 95 watts, where the desktop cards beside it are allowed 350 to 450, does most of the work on the model this workshop runs, loses a third of it on a dense one, and gets the third back the moment it is allowed 150. Every watt on the page is the card's own draw and not the wall's, the fan was held at maximum by hand, and nine things the page cannot say are written down. The bench 2026-09-21 Four homes for one reranker: what RuleSage's slowest stage cost at each stop On the morning of 2026-09-17 an operator at a game table waited nearly twenty seconds for the first word of a rules answer and thought the app was broken. It was not. One small stage of the answer, a cross-encoder that re-reads the shortlist before anything is written, had moved to a rented server whose processors lack the instruction set its weights were built for. This page follows that stage across four machines, three of them in five days: two rented VPSes, a mini PC, and a graphics card that was already in the house. Every millisecond we measured at each stop is here, with the gate we ran before the last move, the first word a player waits for after it, the traps that bit on the way, and what is still unmeasured. The bench 2026-09-21 A 3090, in its own words Fiction, and the byline says who wrote it: a used EVGA GeForce RTX 3090 XC3 Ultra, 24 GB, tells its three lives in the first person: a year of mining, a sky across two monitors, and now the minds it serves. Workshop notes 2026-09-20 Introducing the playground One door, and behind it a rotating, curated set of four exhibits you can walk into. Today: a dinner party for the dead, a drum machine with a genie, a darkroom on a public wall, and a walking tour with footnotes. Workshop notes 2026-09-19 Two 3090s out, one 3090 Ti in Two GeForce RTX 3090s came out of service and one EVGA GeForce RTX 3090 Ti FTW3 Ultra went in; a print lab and a beat lab moved with them on 2026-09-18 (UTC), with nothing closed. Every figure here is either a bench reading from the companion page or a line from a service's own journal, and every row says which. Workshop notes 2026-09-18updated
2026-09-20 The indifferent prediction On the night OpenAI published a resolution of the Navier–Stokes existence and smoothness problem, an operator — one person, the same one throughout this page — and a model looked at the one figure in the post and each said what they saw. Nothing on this page is this workshop's measurement: it is two readings of somebody else's result, the story the picture arrived into, and the thought the night ended on — about a third kind of mind, one that would learn the universe without ever learning us. Workshop notes 2026-09-18updated
2026-09-22 RTX 3090 vs RTX 3090 Ti: an inference and diffusion bench One EVGA GeForce RTX 3090 XC3 Ultra 24 GB and one EVGA GeForce RTX 3090 Ti FTW3 Ultra 24 GB, in the same x16 slot of the same desktop, on the same supply and the same script, 17–18 September 2026. Numbers only: what each writes, what each holds, what a picture costs, how hot each ran, and what a power cap takes away. The bench 2026-09-17updated
2026-09-21 Two 3080s against one 3090, and the cap decides Two GeForce RTX 3080 10 GB cards bought in 2021, in a desktop of the same vintage, against one GeForce RTX 3090 24 GB in the same box, on the same supply, through the same script. The single card is faster on three of the four things measured and uses about half the energy; the pair reads a long document in two thirds of the time and costs about two fifths as much. The bench 2026-09-16updated
2026-10-30 A Short History of Mistral A French lab that put its first model on a torrent, every open-weight release since with its licence, and what its 24B model does on a 3090 and a 96 GB workstation card. Workshop notes 2026-09-16 Ask About This Page Every article on this hub has a room now, where a model on our own machines answers from that page’s own words and leaves its first draft on the page, marked where the checks took something out. This is how it is built, what it keeps, what it refuses, and what it got wrong on the way here. Workshop notes 2026-09-15updated
2026-09-21 A Dinner Party for the Dead A dinner party for three people who are dead and do not know it, hosted by you. Six guests are on the shelf, played by two models on our own machines. The field guide 2026-09-14updated
2026-10-21 The cost to purchase and run a very capable home AI rig A consumer graphics card in a consumer desktop, nothing server-grade anywhere in the box: the rig costs $3,198.86 with one card and $5,048.85 with two on 13 September 2026, before tax. Running it costs about $18 a month with one card and $31 with two if every day puts them to work for six hours; with the cards bought used at August's average completed sale, the rig is $2,443.31 with one and $3,537.75 with two. The bench 2026-09-11updated
2026-09-12 How the print lab works A darkroom that develops three pictures a day for your household, a board that fills up and then finishes for good, and an archive where nobody has a name. The field guide 2026-09-10 The Two-Hour Machine, Priced The computer that kept answering at about twelve tokens a second when this workshop pulled its plug — for a projected two hours on a battery — costs $959.00 today as one listing, or $899.98 in three parts, before tax, from one US retailer's new stock, priced on 10 September 2026 with the listing every line came from. The memory is the second-biggest line of the three-part build. The bench 2026-09-09updated
2026-10-30 The Ceiling Is Not the Corpus We trained a fifth adapter for ACE-Step, our text-to-music engine, on 416 Creative Commons house tracks — twice the last training set, measurably tighter, captioned from the artists’ own tags, the best pass shipped instead of the last — and then one of us listened blind to twenty-four pairs and still preferred the model untouched, on every request that asked for house. The last adapter of the series, and the page says why: the ceiling is the composition, not the corpus. The field guide The bench 2026-09-09 Hear It for Yourself Every clip the house adapter’s run produced — eighty-four renders, each as the raw render, the copy the blind sitting played, and a listening master with one constant gain — with the answer sheet unsealed. One of us sat thirty-six blind presentations, twenty-four pairs of the untouched model against the adapter and twelve where both sides were the same file, and every answer is printed beside its trial, as typed. This page is where you check that verdict. The field guide 2026-09-07 Three New Voices at the Narrator's Chair “Are you staying to watch with me?” — GLM 5.3, speaking as the cove’s nine-year-old — one reply, from one of three new voices — GPT-6 Astra, Claude Fable 5.1 and GLM 5.3 — that sat down this weekend in a fishing town that exists only inside RealKeep, the small role-playing game this workshop builds, and took its narrator’s chair: the seat that speaks as a townsperson to whoever has walked in. Beside those three sat the model that holds that chair for real players today, and every reply below was read blind by judges from outside model families who were never told which voice was whose. The bench 2026-09-06updated
2026-10-01 Two New Frontier Models at the Rules Desk RuleSage, our board-game rules helper, answers a question from the rulebook’s own words: it finds the passage, cites it by number so a player can check, and says plainly when the book does not answer. GPT-6 Astra and Claude Fable 5.1 sat that job’s exam, frozen in August, judged blind by models from seven families that built neither, beside the small local model that already does the job. Neither new model cleared it, and a blind seven-family panel did not separate the two over thirty-six cases. The bench 2026-09-04updated
2026-09-11 Six Worlds for the Print Lab Six worlds, painted large, for a print lab that is now open — a darkroom out back, a wall out front. The gallery 2026-09-04updated
2026-09-18 The Beat Lab, asked and answered Every question the Beat Lab’s genie can answer, answered here too — the same twenty-eight answers, out of the same file, read by a page instead of a genie, each one set beside the sentence of the field guide it was written from. The field guide 2026-09-03 Ten Minutes with Living Artists A genre instead of a composer: 159 tracks of Creative Commons minimal techno, the same ten-minute recipe that taught three dead composers, and the one ear in the house preferred the model with nothing added — both times. The field guide The bench 2026-09-03 Listen for Yourself Every clip of the fourth adapter’s four rounds — sixty-five renders, raw and brought to a common loudness — so you can hear what one listener heard and decide for yourself. The field guide 2026-09-02 Half an Hour with Dead Composers Two more dead composers taught the same way — a hundred and six marches, then Bach — and then blending composers with linear algebra, where the obvious method is wrong, the right one costs a measurable one percent, and a sentence already staged for publication had to be taken back before you ever saw it. The field guide The bench 2026-09-01 Teaching a Music Model Chopin in Five Minutes Nineteen public-domain Chopin recordings, four minutes and forty-nine seconds of training on a five-year-old graphics card, and a dial you can hear — five clips on the page that differ from each other in exactly one way. The field guide The bench 2026-08-31 The Same Sixteen We put glm-5.3-flash — ollama’s new 320B-total cloud model — on two frozen house instruments; it tied our local 12B and 27B at 16/19, and the finding is about the ruler. The bench 2026-08-31updated
2026-10-07 Two Hours at 12 tok/s — On Battery Take the graphics card away and the 30-billion-parameter model answers faster than the 9-billion one — on all three machines. The bench 2026-08-29updated
2026-10-01 The Instrument Travels Carry a frozen exam from a 96 GB workstation card to a 24 GB consumer one and the model it ranked two weeks ago ranks again — at the floor, on a different build, with nothing new seated. The bench 2026-08-28updated
2026-09-04 Nine Worlds, One Dog The founder's dog takes the same nap under the same tree in all nine painted worlds — same dog, same curl; only the world changes. The gallery 2026-08-28updated
2026-10-07 What 150 Watts Buys Turn a 600-watt card down by a quarter and one production model pays two percent of its speed; the other never notices. The bench 2026-08-27updated
2026-09-15 How the Beat Lab works A drum machine that lives in your browser, a genie that runs on our own hardware, and a board where nobody has a name. The field guide 2026-08-27 The Typist and the Developer Two ways a machine writes: one word at a time, or every pixel at once — sixteen times over. The field guide The bench 2026-08-26 A Rig Your Friend Already Owns We re-rendered our game's actual art — same prompts, same seeds, same checkpoint files — on a several-year-old consumer RTX 3090 and a gaming laptop, paired render-for-render against our 96 GB card's own archived output. Two of the three runs happened in the same physical computer. This page is the stopwatch, told honestly — including the one number we don't yet trust. The bench 2026-08-26 How a Vision Model Sees: Eyes for a Machine That Reads A language model reads everything as tokens — little word-pieces, one after another. Before it can say anything about a photograph, the picture itself has to become tokens. This is the close-up on that bridge — what an image becomes on the way in, what it costs, and what a small model actually sees when it looks — with the gate our own game puts every uploaded photo through as the working example, and the gate’s own answers watched against it. The field guide The bench 2026-08-25 Fifty-Seven Milliseconds: What a Diffusion Step Actually Buys One picture, drawn rung by rung up the step dial, on two painters that could not be more different. Timed to the millisecond and measured to the pixel, twice, a month apart — and the two runs say two different things, which turns out to be the finding. The bench 2026-08-25updated
2026-09-04 The Dog, the Dice, and the Painter: How a Small World Draws Itself A photograph of a real dog walks into a painted fishing cove and comes out as a painted dog the world will remember. This is the whole pipeline between those two facts — the file, the dice, and the painter — one station at a time, with no scores and no instrumentation. The field guide

2026-08-24updated
2026-08-24 The new-arrivals diffusion bench One small fishing village, painted thirteen ways — Sorrowmoor Cove (and some other worlds built on RealKeep) sits for thirteen open-weight painters: 3,588 published renders across seventy-five galleries, every hero hand-picked, every licence read first-hand, and the two painters whose licences refuse counted in the open rather than quietly dropped. The bench

2026-08-24 It Was Already There: Where New Knowledge Actually Comes From Models don’t conjure knowledge out of nothing — they read archives nobody had finished reading and walk rooms the rules imply, and what turns either into knowledge is the check that follows, never the confidence of the proposal. The field guide 2026-08-23 It’s Not a Metaphor: A Language Model Is a Compression of Its Corpus A language model is a compression of its corpus — not a figure of speech but the training objective, and once you hold that one fact, bias, hallucination, scale and the frozen moment stop being mysteries and start being mechanics. The field guide 2026-08-22 The free speed wasn’t free — the runtime dividend, traced to one commit One day our busiest model started producing tokens roughly a third faster than it ever had, and nobody had touched it. The cause was a two-line upstream commit ending a ~2.5 ms-per-token tax nobody configured — and, in the same stroke, changing what every unpinned model says. The receipts, the mechanism, and the honest unmeasured half. The bench Workshop notes 2026-08-21 The compressed photograph — what those Q4_K_M tags actually mean Nearly every local model wears a tag like Q4_K_M, and most of us read one thing from it: fewer bits, smaller file. We opened all four builds the vendor ships of one 11.9B — 4.80 to 16.13 bits per weight — and asked each the same question: the 8-bit answer is byte-identical to the uncompressed original’s, and the wrong answer survives the whole ladder. The field guide 2026-08-20updated
2026-10-21 Everyone on the payroll, three at the table — dense, MoE, and active weights Model names grew a second number — 30b-a3b — and it is the one your card and wallet care about. The payroll and the meeting, in plain English, measured on our own machines: a 25.8B mixture out-writing its 11.9B dense sibling while holding twice the memory — and the robber asks both architectures. The field guide 2026-08-19updated
2026-08-21 Reading is fast, writing is slow — why AI answers arrive word by word Paste a whole essay into a chatbot and it swallows the thing at once — then answers one word at a time. That asymmetry is physics, not theater: reading and writing in plain english, measured on our own card — and the measurement argued back, which turned out to be the best part. The field guide 2026-08-18updated
2026-10-21 Three librarians and a careful reader — how RuleSage finds the right page Before RuleSage answers, something has to find the right page in a book the size of a small novel — three searchers, one referee, one careful reader, and none of them is the model that writes your answer. The whole journey, in plain english, with the shipped numbers. The field guide Workshop notes 2026-08-17updated
2026-08-20 qwen3.8:27b across four house benches: one seat filled, one floor missed qwen3.8:27b landed on our box the day after it shipped, and we sat it in the exams we already had rather than designing anything for it — it filled a screening seat that had never been filled, and it missed a floor in the judge seat. The bench 2026-08-17updated
2026-10-21 The map nobody picks up — and whether it helps when it arrives. Two camps have argued about a map file for machines for two years, mostly without receipts — no AI crawler asked us for ours in thirty days of logs, so we measured whether it helps when it arrives anyway. The bench Workshop notes 2026-08-15updated
2026-08-20 The open call — a kid, an elder, and a tired parent walk into the cove. Twenty language models were handed the same frozen moment of a small fishing town and asked to speak as its people, then read blind by seven judges from six rival families. The bench Workshop notes 2026-08-15updated
2026-11-01 Where your question goes — a plain-english walk. Our two published privacy promises, unpacked for people — not programmers — by following one question door to door. Workshop notes 2026-08-14updated
2026-09-10 How we work — who writes this, and how. The honest answer to the fair question: the workflow that makes every exhibit — humans and agents, named plainly, with the receipts. Workshop notes 2026-08-14updated
2026-08-20 The move — the same yardstick, before and after. Our apps left the gaming laptop for a real server — the same frozen bench measured both eras, honestly. The bench Workshop notes 2026-08-13updated
2026-10-01 The chair trials — five fresh exams for the thirty-billion class. Nine models, eleven arms, five all-new exams in one day — on a box that never stopped serving its real users. The bench 2026-08-13 The narrator’s chair, refused. The August class’s best voice sat the cove’s actual chair — and was refused at the gate. The bench 2026-08-13updated
2026-08-14 The outside judges — four rival labs re-check our work. Four rival frontier labs re-judged every sealed round we had published — blind, byte-identical. The bench 2026-08-12updated
2026-08-20 The August arrivals — the fresh class sits the house exams. Four open-weight models landed in one week, so we gave them the two exams we already had — one frozen since July, one rebuilt in its shape. The bench 2026-08-11 The diffusion research bench. What a fully local image pipeline delivers for a living world — head-to-heads, ladders, cold starts. The bench 2026-08-11updated
2026-08-19 The seat trials — how a local model earns a chair. Before a model earns a seat inside our products, it sits this trial. The bench 2026-08-11updated
2026-08-13 The voice trials — how the narrator earned its voice. Twenty-four model tags across three hardware eras, hunting a voice that can inhabit a character. The bench 2026-08-11updated
2026-10-24 The licence ledger — every model, read first-hand. Every model we run or bench, its licence read from the actual text — the page the others cite instead of restating. Workshop notes

The roster

Looking for one model rather than one bench? The model roster lists every model that has appeared in a published exhibit here — Muse Glimmer, Qwen 3.6 and 3.8, Gemma 4, Nemotron 3.5 Lightning, OLMo 3.1, Granite 4.1, Kimi K3, DeepSeek V4, Mistral Large 3, GPT-5.5, Claude Opus 5, GPT-6 Astra, Claude Fable 5.1, the FLUX.1 painters and the rest — with the exhibits each one appears in, that round’s own verdict quoted word for word, and its row in the licence ledger. It is an index, not a bench: it is regenerated from the exhibits’ data kits and settles nothing on its own.

The lab protocol

Every exhibit on this hub was produced under the same discipline. None of it is a standard anyone handed us — we wrote each law down after a bench told us something we didn't want to hear, and we've kept to them since:

Quality gate first — the pass bar is written down before the experiment runs.
Measured verdicts — a claim without a receipt is not published; every figure is a count of rows in a named artifact, and raw rows go out on request.
Self-refutation honored — when the data kills our favorite, the favorite dies.
Real content — benches run on the products' own material, never toy prompts.
Losses published — a rejected bet gets its numbers shown, not a memory hole.
Nothing leaves the box unnamed — benches run on our own authored content, on our own hardware. When a bench reaches a hosted endpoint (the seat trials' reference ceiling does), the page says so plainly, and no customer or user data rides in any prompt.
Hardware by class — machines are named by VRAM class, not model; within-rig comparisons are exact, cross-rig extrapolation is approximate. Specific specs are available on request — drop a line.