Free tools
The Lab: 5 free tools to choose, cost and secure your AI agents
The Lab brings together five free tools that answer the practical questions before adopting an AI agent: which model is best for its price, what usage will cost, what hardware a local model needs, which agent fits your case and how safe your setup is. All rely on the directory data, collected on October 25, 2026.
LLM benchmark
Which model gives the best score for its price? Interactive ranking of 164 LLMs: Epoch AI scores (GPQA, SWE-bench, ARC-AGI…), OpenRouter prices, filters and a value chart. Data from October 25, 2026. 356 priced modelsLLM price calculator
How much will my agent cost each month? Estimate the real cost of 356 AI models: token counter, prompt caching, cost per task, per day and per month, cheapest first. Prices from October 25, 2026. graphics cards and MacsVRAM calculator
Can my machine run this model locally? How much VRAM for a local LLM? Quantized weights (GGUF), KV cache by context, overhead: a verdict for 15 graphics cards and Apple silicon Macs. 388 agents comparedWhich AI agent should you choose?
Which agent fits my level, my task and my budget? Level, task, budget, hardware, confidentiality: 5 questions to find 3 suitable AI agents among 388, with cost, risk and setup guide. 20 questions, score out of 100AI agent security check
Is my setup safe enough? Sandbox, action approval, secrets, MCP servers, prompt injection: 20 questions for a risk score out of 100 and priority fixes.In what order to use them
If you are starting from scratch, begin with Which AI agent should you choose?: five questions are enough to get three agents suited to your level, task, budget, hardware and data confidentiality. Each suggested agent links to its profile and a guide to get started.
The chosen agent needs a language model. The LLM benchmark compares the scores published by Epoch AI and plots each model on a score/price chart; the price calculator then turns your real usage into a bill per task, per day and per month. If you prefer a local model, for confidentiality or to pay nothing per use, the VRAM calculator tells you which graphics card or Mac you need.
Finally, once the agent is installed, the security check reviews twenty protections, from sandboxing to MCP servers, and ranks the fixes from most to least urgent. Run it again after each new access or extension.
Rules shared by all tools
- Public, dated data. Epoch AI scores, OpenRouter prices, Hugging Face model cards, GitHub repositories: each tool cites its sources and the collection date. No figure is invented or added for show.
- Published methods. Formulas, weights and ranking rules are described under each tool, with their limits, so you can check or challenge the result.
- Nothing leaves your browser. No account, no email requested, no server call during the calculation.
- No paid rankings. No vendor pays to appear or rank higher in a result. Details on the method page.
Frequently asked questions
Are the Lab tools free?
Yes, with no sign-up and no usage limit. They ask for no email and no account.
Is my data sent anywhere?
No. The data needed is included in each page and all calculations run in your browser: text pasted into the token counter or your security-check answers never leave your device.
Are the figures up to date?
They come from the latest collection, dated October 25, 2026, and are refreshed with each new collection. The date is shown at the top of each tool.
Neutrality. No affiliate links. No vendor pays to be reviewed, rated or ranked. Reviews are directed by François Chollet, neutral and factual, and every data point is dated and sourced. How we score