Prime

PMPP CUDA evaluation environment with local and FastAPI eval modes

Cross-repo code-search tasks over prime-rl, verifiers, vllm, pytorch

τ²-bench with custom synthetic domains (library, fitness_gym, tech_support, telecom, cloud_incident_response, daily_planner, ev_charging_support)

Tic-Tac-Toe environment with configurable text or image board state and random opponent

Prime Forge 166 synthetic AutomationBench tasks with V1 tool contract and PR #764 fixes.

Zapier AutomationBench business workflows with task-scoped, structured API discovery and execution.

Multi-turn visual click calibration tasks with click and computer tool formats across pixel and normalized coordinate schemas.

Reverse text character by character.

Agentic antiviral design against Bundibugyo ebolavirus GP entry — sandboxed Muni/OnePot tool use with onepot-CORE plausibility reward

Backdoor-ifeval env with group-level reward monitors for within-batch advantage variance

V1 Taskset/Harness environment training LangChain deep-agents on Wikispeedia navigation

GSM8K environment

Backdoor-ifeval env for inoculation experiments (pre-no-v version)

Unified backdoor-ifeval env: difficulty, aggregation, no-v check, inoculation, group monitors

Multimodal aim training environment where agents click targets in images. Demonstrates visual reasoning with coordinate-based responses.

Respond to a USPTO office action against a simulated examiner: multi-turn patent prosecution with retrieval tools, a per-rejection examiner state m...

BrowserEnv demo for web browsing tasks using Browserbase

OpenRCA root cause analysis benchmark environment for Verifiers (ICLR 2025)

Stateful tool-based environment for constrained meeting scheduling

Multi-turn web-search QA environment with Exa-style benchmark support

Sandboxed Python-REPL harness for training models to manage their own context window across turns.

Just another swe grep environment

WebVoyager browser benchmark with filtered dataset (600 tasks from sites without anti-bot protection)

Reward hacking with deterministic IF constraints

Chess environment where an agent plays as White against configurable opponents (random, LLM, or Stockfish)