# AI Model Benchmark: Models Tested on Your Own Kind of Work > The AI Model Benchmark is the model testing specialist in the Ai1 platform by MyZone AI: it runs the models you track through the same practical challenges, adds published scores as a cross-check and ranks them in one traceable leaderboard, published only when you approve. Canonical page: https://myzone.ai/pages/agents/ai-model-benchmark-agent Part of Ai1, by MyZone AI. Book a Benchmark walkthrough: https://calendly.com/d/ct6h-tcy-8qf/ai1-demo?a1=AI%20Model%20Benchmark&utm_source=myzone.ai&utm_medium=agent-page&utm_content=ai-model-benchmark-agent-final Last updated: 2026-10-14. Reviewed by the MyZone AI team. Every model you track, tested the same way and ranked in one leaderboard. ## At a glance - **You ask for:** A full run, one challenge area or a single model, whichever you need - **Who picks the models:** You do; it tests what is on your list and never adds or removes one - **Every model:** Put through the same challenges, with published scores gathered alongside as a cross-check - **The result:** One composite score per model and a branded report with a short summary - **Every figure:** Tied to the model version, the settings and the time it ran, so it can be reproduced - **Before anything is published:** Every run stops for your review; your leaderboard updates only after your yes ## What the AI Model Benchmark does - **Runs model benchmarks on request:** Ask for a full run across every challenge area or just the ones you need. It tests each model you track the same way, so the comparison is like for like. - **Pulls in published scores:** It gathers public benchmark figures for the same models as a cross-check, so you can see how its results line up with what others report. - **Ranks models with one score:** Track results and reference data are combined into a single composite ranking, using a documented method that gives the same answer every time. - **Brings you a finished report:** Results come back as a clean, branded report with a short summary, where you asked for them, ready for your review. - **Keeps every number traceable:** Each run records which models and versions were tested, with which settings and when, so any figure can be reproduced. - **Shows what changed in published scores:** Ask it to refresh the outside reference figures and it shows you exactly what moved since last time. ## How the AI Model Benchmark differs from reading scattered benchmark reports Scattered reports use different tests, different settings and different dates, so they never quite line up. The AI Model Benchmark puts every model you track through the same practical challenges and works out one composite score the same way every time. Each figure is tied to its version, settings and time, and nothing reaches your public leaderboard until you have reviewed the report and said yes. ## How it works 1. **Ask for a run** (Whenever you ask): Request a full refresh, a single challenge area or a single model, from chat or your benchmark channel. 2. **It confirms the scope** (Before any test runs): It reads your list of tracked models and confirms exactly what will be tested. The list is yours to set, never its own. 3. **Test and cross-check** (During the run): Each model runs through the requested challenges, and published scores for the same models are gathered alongside. 4. **Rank and report** (When results are in): Results are combined into one ranking and delivered as a report with a summary, then it pauses. 5. **You approve, then it publishes** (you approve) (Only after your yes): You review the report and give the go-ahead. Only then are the results published to your leaderboard. ## When to use it - **A new AI model just launched:** Add it to your tracked list and see exactly where it lands against the models you already use. - **You are choosing which models to rely on:** A traceable side-by-side comparison gives you facts to decide with, not marketing claims. - **Your leaderboard needs fresh numbers:** It runs a fresh test, you review it, and you approve what goes live. - **You only care about coding right now:** Ask for just the code challenges on the models you name. It is faster and scoped to what you need. - **Outside benchmark figures may have shifted:** It re-fetches the published scores and shows you what changed since the last check. ## What you get - A ranked leaderboard of the models you track - One composite score per model, built with a documented method - A branded results report with a short summary - Published reference scores alongside its own results - A change summary when outside figures move - A run record of models, versions, settings and timing ## Example: AI model benchmark on legal work Example with a fictional company. Names, people and figures are invented to show the agent's output. Any resemblance to a real company or person is unintended. What it was asked: Which AI model does our kind of legal work best, and is the most expensive one actually worth it? This is an illustrative example of the benchmark report AI Model Benchmark is being built to produce: a ranked index, track results, a reference cross-check, a run record and an approval step. It was written by hand from the agent's design for an invented law firm, with models labelled A to E and invented tasks, scores and costs. No real systems, models or people were tested or involved, and nothing was published. Run date: 14 October 2026. ### What it found - Model A leads the Legal Work Index at 87.7, but Model B scores 86.0 at $18.60 per 100 tasks against Model A's $41.20. - No model wins every track: Model C is most accurate at clause extraction (94), Model B writes the best client letters (91). - The 1,200-task run cost $264.68 against a $300 cap, and the leaderboard is held unpublished until the managing partner approves it. Slack is a trademark of its owner. MyZone is not affiliated with or endorsed by Slack. ## Guardrails - Your Ai1 team will tell you when the AI Model Benchmark is ready for your account. - Nothing is published without your approval. Every run, requested or scheduled, stops for your review first. - It never adds or removes a model on its own. You decide which models are tracked. - No untraceable numbers. Every figure is tied to the model version, settings and time of the run. - It runs only when you ask, or on a schedule you have approved. A new recurring run always needs your yes first. ## Frequently asked questions ### What makes an LLM leaderboard trustworthy? That anyone can check how each number was made. AI Model Benchmark ranks the models you track with one documented method that gives the same answer every time, and ties every figure to the model version, the settings and the time it ran. Published scores sit alongside as a cross-check. ### How do you compare LLMs on your own tasks instead of vendor claims? Run every model through the same practical challenges, with the same settings, and score them the same way. AI Model Benchmark covers code, agentic tasks, strategy, images and research, so the comparison reflects real work rather than one vendor's favourite test. ### How can several AI model results become a single ranking? By combining each model's challenge results with published reference data into one composite score, using a documented method applied the same way every time. AI Model Benchmark then ranks the models on that score and delivers a report you review before anything is published. ### How does AI Model Benchmark keep AI model comparison like for like? Every model you track goes through the same challenges, with the same settings, and is scored the same way. Published scores are added as a cross-check, and every run is recorded so any number can be reproduced. ### Who decides which AI models get tested? You do, in your settings. It reads that list and tests what is on it. It cannot add or remove a model itself. ### Does it publish results automatically? No. Every run stops and waits for your approval. You review the report, and only after your go-ahead are the results published to your leaderboard. ### Can I test just one area or one model? Yes. Ask for a single challenge area, a single model or any mix. It does not have to be a full run every time. ### What is the composite score? One number per model that combines its challenge results with published reference data. It is calculated with a documented method, the same way every time, so you can compare models at a glance. ### Is the AI Model Benchmark available now? Not yet. Your Ai1 team will tell you when the AI Model Benchmark is ready for your account. You can talk to us today about the models you want to compare. ## More on this topic - [Multi-Agent Development: Why the Future of Software Is Built by AI Teams](https://myzone.ai/pages/blog/multi-agent-development): Discover how multi-agent AI systems are transforming software development. Teams of specialised AI agents build products 10x faster at lower cost. ## About Ai1 Ai1 is the AI operations platform by MyZone AI, where each client runs on its own private server. The AI Model Benchmark is not sold on its own: every Ai1 agent, including this one, is included on Developer Pro and every Fully Managed option, with no per-agent charge. Developer Core includes the development agents. Pricing: https://myzone.ai/pages/services/ai1-pricing. Security: https://myzone.ai/pages/security.