verifiers

What it is and what it does

Verifiers is a library for creating and managing training and evaluation environments for large language models, built around reinforcement learning workflows. It is designed to work within the Prime ecosystem—specifically the Environments Hub, the prime-rl training framework, and the Hosted Training platform—but can also be used as a standalone harness for multi-turn agent interactions, tool-use training, and LLM verification tasks.

The package brings together environment management, evaluation harnesses, and agent integration under a single interface. It depends on a substantial set of libraries for async I/O (aiohttp, httpx), LLM provider clients (anthropic, openai), data handling (datasets, numpy, pydantic), and distributed compute (pyzmq, uvloop). The library is actively maintained, recently released, and targets Python 3.11–3.13.

Use it for:

• Train agents using reinforcement learning (GRPO, agentic RL) with multi-turn environments and tool-use capabilities.
• Evaluate LLM outputs against verification criteria in a structured harness integrated with the Prime platform.
• Build custom evaluation environments for agent reasoning and decision-making tasks.
• Integrate LLM training pipelines with the Prime CLI and Environments Hub for collaborative model development.
• Benchmark multi-turn agent behavior in controlled, reproducible environments.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Verifiers provides environments and evaluation harnesses for training and assessing large language models using reinforcement learning, with integration into the Prime ecosystem.

Yes, if you are training or evaluating LLMs with reinforcement learning and plan to use the Prime ecosystem. The package is actively maintained, has no known vulnerabilities, and integrates tightly with prime-rl and Hosted Training. If you need a standalone environment harness without Prime integration, verify that the 26 dependencies and ecosystem lock-in align with your workflow first.

Install

verifiers on PyPI

pip install verifiers

uv add verifiers

poetry add verifiers

Installing verifiers

Before you install

Low friction installation with a pure-Python wheel. Active maintenance—released 7 days ago with 4513 repository stars. Requires Python 3.11–3.13 and pulls 26 runtime dependencies including aiohttp, anthropic, openai, and pydantic, which are standard in the LLM ecosystem.

License in practice

MIT license (permissive). No restrictions on commercial or private use; you may modify and redistribute under the same license.

Quickstart

pip install verifiers

from verifiers import Environment
env = Environment()

Requires Python 3.11 or later. The package is tightly integrated with Prime CLI and the Environments Hub; standalone use may require additional setup.

Verify before relying

• Specific API surface and core classes beyond Environment—documentation depth unclear from excerpt.
• Whether 26 runtime dependencies are all required or some are optional/conditional.
• Integration requirements with prime-rl and Hosted Training platform for full functionality.

Package facts

Evidence: verifiers-0.3.0-py3-none-any.whl

Keywords: agentic-rl, agents, environments, eval, grpo, harness, llm, multi-turn, reinforcement-learning, rl, rlvr, tool-use, train, verifiers

Tags

More Python Modules packages

Further reading