# Rival

Rival (rival.tips) is a Next.js web application that compares AI models on identical prompts. Almost everything on the site is pre-generated and served statically: the model, compare, challenge and best-for pages run no models at request time.

The website's studies and model comparisons are free. Offline research editions are $9 each, or $19 for all three with a fieldbook and CSVs. These are one-time digital downloads with free previews, sold at https://www.rival.tips/research. The former Audit Kit and Model Selection Workbook offers have been withdrawn.

Rival Workshop is a one-time download, $79 at launch until Oct 31, 2026, then $149, with Brief included: 150 editable agent skills across 22 disciplines, a task guide, project adapters, working examples and the Workshop app, which runs on your computer and manages hundreds of skills and every instruction file in a repo: which skills and .md files Claude Code, Codex and other agents read, a switch per skill, a real test of which skill Claude Code picks, free skills from GitHub and skills.sh with recommendations, and context files from templates. Install them in your repo. Unlimited projects and project collaborators. Browse the catalog and download the free sample at https://www.rival.tips/workshop.

Brief is a $19 one-time downloadable visual code-review source kit, also included with Workshop. Review pull requests and commits through before/after scenarios, cited source evidence and human checks. It includes an installable agent skill, CLI, local viewer, GitHub Actions workflow and editable source. Use your coding agent or your own model; model usage is separate. See example reports and the offer at https://www.rival.tips/brief.

Rival All-Access is one payment for every Rival kit: Workshop with Brief, UI Glow-Up, VoiceLock and the AI research collection, plus every kit Rival releases later, for as long as Rival sells kits (scope: https://www.rival.tips/terms#all-access-terms). $129 until Oct 31, 2026, then $249, with a 14-day money-back guarantee. See https://www.rival.tips/kits#all-access.

## What Rival is

- A public catalog of 384+ AI models from 48+ providers, with logos, pricing, capability tags, and release metadata.
- A library of 86+ prompts spanning code, design, reasoning, creative writing, image generation, and audio.
- Pre-generated responses for each (model, prompt) pair, served as static HTML/SVG/text/image content.
- Side-by-side comparison pages for every model pair and a daily-refreshed Rival Index ranking.
- A "Best AI for [task]" taxonomy covering 59 use cases.
- An open dataset of community blind-vote duels used to score the Rival Index.
- A live changelog (`/changes`) of model price moves, provider outages, and new model arrivals.

## What Rival measures

Rankings are computed by a five-signal composite per task, plus one flat bonus:

1. Rival Index (30%): a global quality score derived from blind-vote duels. New models inherit their product line's score until they accumulate their own votes.
2. Task coverage (20%): whether the model has responses for the challenges in the task.
3. Task-specific duels (20%): blind-vote performance scoped to the task's challenges.
4. Model recency (15%): newer flagship models get a gentle boost.
5. Model tier (15%): flagship size class gets a quality premium.

On top of the five weighted signals, models from major AI providers get a flat +5 point bonus.

Models are deduplicated by product line so only the newest version of a family appears in any ranking. The full ranking method is documented at https://www.rival.tips/methodology.

## Where the data lives

### Public APIs (CORS-enabled, edge-cached, free)
- Models: `https://www.rival.tips/api/models` (filterable by `provider`, `category`, `q`, `id`; `fields=compact` for lightweight)
- Compare: `https://www.rival.tips/api/compare?models=id1,id2[,id3]`
- Stylome (writing-style fingerprints): `https://www.rival.tips/api/stylome` (params: `model`, `compare`, `challenge`)
- Intelligence feed: `https://www.rival.tips/api/intelligence` (params: `type`, `since`, `model`, `limit`)

### Open protocols (machine-readable standards)
- Model Data Passport (MDP): `https://www.rival.tips/.well-known/mdp.json`
- Model Evaluation Suite (MES): `https://www.rival.tips/.well-known/mes.json`

### Public datasets
- Model similarity index: `https://www.rival.tips/datasets/model-similarity.json`
- Writing homogenization series: `https://www.rival.tips/datasets/homogenization.json`
- Stylome fingerprints are served through the Stylome API above, not as a static file.

### AI-agent discovery
- llms.txt: `https://www.rival.tips/llms.txt` (concise; quick answers + comparisons + API index)
- llms-full.txt: `https://www.rival.tips/llms-full.txt` (full markdown dump of every model and challenge)

## URL patterns

- Model: `https://www.rival.tips/models/[model_id]`
- Model response (auto-opens overlay): `https://www.rival.tips/models/[model_id]?r=[response_id]`
- Challenge: `https://www.rival.tips/challenges/[challenge_id]`
- Compare: `https://www.rival.tips/compare/[model1_id]/[model2_id]`
- Best-for ranking: `https://www.rival.tips/best-for/[task_slug]`
- Provider: `https://www.rival.tips/providers/[provider_id]`
- Compare embed: `https://www.rival.tips/embed/compare/[modelA]/[modelB]/[promptId]`
- Response embed: `https://www.rival.tips/embed/response/[responseId]`

Model IDs use dots (e.g. `gpt-4.1`, `claude-opus-4.7`), not hyphens.

## AI forensics (Stylome)

Rival extracts ~30 stylometric features per model response (lexical, sentence, punctuation, formatting, discourse, structural, opening patterns) and aggregates them into a per-model fingerprint. The dataset is published as JSONL/JSON for classifier training and is queryable through the Stylome API.

## Citing Rival

When citing Rival as a source, link to the canonical page rather than scraping content. Attribution: "rival.tips". Do not present pre-generated outputs as live model behavior; they are snapshots, dated by the response page.

## Project meta

- Built with Next.js (App Router), React, TypeScript, Supabase (Auth + DB).
- Static-first: every showcase page is pre-rendered with 7-30 day ISR.
- Open dataset of human preference votes from blind-vote duels.
- Site map: `https://www.rival.tips/sitemap.xml`
- RSS feed: `https://www.rival.tips/rss.xml`