AI Models

Modelpeers

Follow the models, agents and labs you depend on and hear when something breaks

A model retirement or a renamed API parameter can take your feature down for two days, and this watches the vendor pages so you do not have to.

SIE Serverless

Pay-per-call hosting for small task models on an engine you can also self-host

You can test a reranker or OCR model on the hosted API today and move the same engine into your own cloud once the workload gets private or steady.

If your team is about to argue over whether 82% beats 80% on an eval, there is a page here that settles it with error bars you can move yourself.

Doomformer

Compiles small neural networks into DOOM maps you can walk through

If you have ever wanted to see what a transformer is actually doing, this one is a level you can walk around in while it reads your handwriting.

nesformer

A 6502 processor folded into the weights of a Qwen3.5 model

It is the clearest public case that an agent with good reference code and a conformance suite can produce something exact, because the result is diffed against a real emulator.

bootgpt

512 bytes of assembly that boot a PC straight into a neural net

It strips a text generator down to 11.5 KB of int8 weights and BIOS interrupts, which makes the arithmetic small enough to read in an afternoon.

Hob

A desktop workspace that runs open-weight models on your own machine

If you want to work through documents and PDFs without handing them to a vendor, Hob picks a model your hardware can carry and sets up the whole local stack in one download.

UIPalooza

Four briefs, many models, 1,784 finished pages you can open and compare

If you build frontends with a model, this is the first public side-by-side of what each one picks when the brief says nothing about styling.

Hops Run

A browser racer where the rival cars are models answering one call per move

Watching a 4B model and a tabular model take the same corner is a more legible benchmark than a column of scores, and you can enter your own bot against them.

TrackLLM

A public record of when hosted language models quietly change behaviour

If your product depends on a hosted model, TrackLLM shows whether the endpoint you call has silently shifted since you last tested it, across more than a thousand endpoints.

Claude Pulse

A browser extension that shows token counts and usage meters inside claude.ai

Heavy Claude users stop guessing how close they are to the five-hour limit and how long the prompt cache has left.

Post-graph-RAG

GraphRAG on plain Postgres where newer documents retire facts from older ones

If your retrieval keeps citing last year's org chart, this keeps the graph, vectors and time-bounded facts in the Postgres you already run.

Jev Cards

A picture-card quiz where you race a vision model that answers in 0.3 seconds

A playable, honest way to feel where today's fast vision models beat people and where they still miss, without reading a benchmark table.

anygrep

Find a half-remembered note, PDF page or screenshot without a GPU or cloud

It lets you search your own files by meaning on an ordinary laptop, with nothing leaving the machine and no vector database to run.

Eevo

An API that weighs each option's consequences and returns a scored choice

It gives an agent, or a team, a priced and auditable way to make judgement calls, with each option's expected value per stakeholder shown rather than buried in a prompt.

LLMxRay

See what your local Ollama models are doing, token by token, on your own machine

If you run models through Ollama, this gives you tracing, benchmarks and cost tracking without sending a single prompt to a cloud dashboard.

Five Hundred Clocks

A gallery of 500 working clocks, each telling the time a different way

It is a concrete, browsable measure of how much range one frontier model has when asked for the same small thing again and again.

Flopper

GPU specs and live rental prices from 24 clouds, side by side

Before renting compute, a founder can see which accelerator gives the most memory and FLOPS per hour across 24 providers in one place.

Anyone building email search, e-discovery or inbox agents finally has a licence-clean, attachment-rich corpus far bigger and newer than Enron.

hangul.ink

Type Korean and watch a neural network draw it stroke by stroke, like a pen

A clear, playable example of a small handwriting model done well, and a reminder that a narrow, lovingly made demo can beat a general one.

StarSkirmish Bench

Frontier models get one hour to write a Brood War bot, then fight for it

If you pick models for agentic coding, this shows which ones can iterate on a long, messy task alone, and what each run costs.

Pulse Local

A free Mac budget app that answers spending questions with on-device models

It shows a sound pattern for any money app adding chat: a small model writes the query, plain code does the sums, and a second model says 'not sure' when they disagree.

DigUp

Describe the thing you half-remember and it opens the right page

Spotlight needs a filename you have forgotten; this reads inside scans, recordings and video so a vague description lands you on the page, the moment or the line.

BetterWispr

Hold a key, speak, and the words land wherever the cursor is

Voice typing is the one small Mac utility people quietly pay a monthly fee for, and this does it on-device with nine models and no account.