// category · 21 tools · 17 open source
LLM Evaluation Tools
Frameworks and platforms for testing language models and LLM applications, from academic benchmark harnesses to CI friendly unit tests, LLM as judge scoring and red teaming kits.
LLM evaluation tools test how well a language model application does its job before and after it ships. This category covers open source frameworks, hosted evaluation platforms and red teaming kits. They differ from public benchmarks: instead of ranking models on a shared exam, they score your own prompts, your own retrieval pipeline and your own agent against test cases you define.
Evaluation frameworks. DeepEval, Ragas, promptfoo, OpenAI Evals, Inspect AI and the LM Evaluation Harness let you write tests in code and run them in CI. Some focus on retrieval quality, with metrics such as faithfulness and context precision; others are built for running academic benchmarks against many models.
Hosted platforms. Braintrust, Patronus AI and similar services add datasets, experiment tracking and dashboards on top of scoring, which helps teams compare prompt and model changes over time.
Red teaming. garak, PyRIT, DeepTeam and Giskard probe models for jailbreaks, prompt injection, data leaks and other failure modes. They complement normal quality tests rather than replacing them.
Most modern evals rely on three kinds of scorer. Code checks test exact or structured outputs. Reference metrics compare an answer with a known good one. LLM as a judge uses a second model to grade answers against a rubric. Judges scale well but show position and verbosity biases, so validate them against human labels on a sample first. Narrow pass or fail questions tend to be easier to validate than vague one to five scales.
A sound process looks like this: collect real examples, write clear criteria, build a test set that includes known failures, run it on every change, and add production failures back into the set.
Start with our playbook on how to evaluate LLM applications, then read LLM as a judge, RAG evaluation metrics and best LLM evaluation frameworks. For security testing, see LLM red teaming tools.
// guides
Guides to LLM evaluation tools
Best LLM Evaluation Frameworks in 2026, Compared
The best LLM evaluation frameworks in 2026: Promptfoo, DeepEval, Ragas, Inspect and TruLens for apps, plus lm-evaluation-harness and HELM for model benchmarks.
LLM Red Teaming Tools: Promptfoo, garak, PyRIT
The best LLM red teaming tools in 2026 compared: Promptfoo, garak, PyRIT, DeepTeam and Giskard, with licenses, first commands and OWASP LLM Top 10 coverage.
Inspect AI vs DeepEval: Which Should You Use?
Inspect AI vs DeepEval in 2026: how the two open source eval frameworks differ on who runs the model, scoring, agents, sandboxes, CI and which team each one suits.
OpenAI Evals vs DeepEval: Which to Use?
OpenAI Evals vs DeepEval in 2026: platform shutdown timeline, DeepEval's pytest CI metrics, multiprovider support, and Promptfoo as the migration path.
LLM Evaluation Metrics: What to Measure and How
The LLM evaluation metrics that matter in 2026: exact match, BLEU, ROUGE, BERTScore, pass@k, LLM judge rubrics, RAG and safety metrics, and when to use each.
How to Evaluate LLMs Before Production
How to evaluate LLMs before production: set the product decision, build a production like offline suite, wire CI gates, then open shadow or A/B traffic.
How to Evaluate LLM Model Performance
How to evaluate LLM model performance before you pick one: build a task test set, fix settings, add error bars, test prompt sensitivity, then weigh cost.
LLM Hallucination Detection: Methods and Tools
How LLM hallucination detection works in 2026: faithfulness checks, detector models like HHEM, self consistency and semantic entropy, and the tools to use.
How to Evaluate LLM Applications (2026 Guide)
How to evaluate LLM applications step by step: error analysis, code checks, calibrated LLM judges, CI gates and production sampling, with tools for each stage.
Prompt Injection Attacks: Examples and How to Test
What a prompt injection attack is, direct and indirect examples, why there is no complete fix, and how to test your LLM app with Promptfoo, garak and PyRIT.
Prompt Injection Detection Library: What to Use
Which prompt injection detection library to use in 2026: Llama Prompt Guard 2, Protect AI DeBERTa, Pytector, rule scanners and APIs, with our own 12 prompt test.
How to Use the PyRIT Tool: A Tested Walkthrough
How to use the PyRIT tool from Microsoft: install it, configure a target, run your first attack with converters and a scorer, then try the scanner and Crescendo.
Open LLM Leaderboard v2: How to Use It in 2026
Open LLM Leaderboard v2 explained: its six benchmarks, how normalized scores work, why Hugging Face retired it, and how to query or rerun it today.
LLM as a Judge: How It Works and When to Trust It
LLM as a judge explained: how model graders work, the biases that skew them, how to validate one against human labels, and which tools to use.
LLM Judge Model: Frontier, Small or Decision Model?
Which LLM judge model should grade your outputs? Frontier LLMs, small judges and decision models like Jev compared on accuracy, cost and calibration.
Patronus AI Agent Failure Research, Explained
Patronus AI agent failure research in plain English: what the 63% figure really means, what the TRAIL benchmark found and how to find failures in your own agents.
How to Use Promptfoo: A Tested Walkthrough
How to use Promptfoo, step by step: install it, write a promptfooconfig.yaml, add assertions, compare two versions, read results and gate CI on the pass rate.
What Is Answer Completeness in RAG Evaluation?
What is answer completeness in RAG evaluation? The share of key facts from a reference answer that a response covers, and how Azure, Bedrock and Ragas score it.
How to Use the MMLU Benchmark, Step by Step
How to use MMLU benchmark runs: pick the right variant, run it with lm-evaluation-harness or Inspect, smoke test the format, and read the score honestly.
What Is Faithfulness in RAG Evaluation?
What is faithfulness in RAG evaluation? The share of an answer's claims backed by the retrieved context, and how Ragas, DeepEval, TruLens and HHEM score it.
RAG Evaluation Metrics: Ragas vs DeepEval vs TruLens
RAG evaluation metrics explained: faithfulness, relevancy, context precision and recall, which need a reference, and how Ragas, DeepEval and TruLens differ.
Jev Decision Index: How Decision Models Rank
The Jev Decision Index ranks 275 decision models on private tests. How the Full score works, who leads, why Jev sits 55th, and how it differs from AA's index.
LLM as a Judge vs Agent as a Judge: When to Switch
LLM as a judge vs agent as a judge: what each grades, what DevAI, AJ-Bench and Mind2Web 2 found, what an agent judge costs, and when to switch.
Ragas vs Langfuse for RAG Evaluation (2026)
Ragas vs Langfuse is a metrics library versus an open source tracing platform. How scores attach to traces, when to use both, pricing, and a tested wiring note.