TL;DR
This guide develops Every RL Benchmark, Explained: Atari → MuJoCo → SWE-bench → WebArena → GAIA → OSWorld from its central engineering problem to practical design choices and limitations. It synthesizes the roadmap topics without claiming new experimental results. Reported findings belong to the cited sources.
Target searches: “RL benchmarks”, “agent benchmarks explained”
Why This Topic Matters
Modern LLM and agent training turns familiar reinforcement-learning ideas into systems problems: actions may be tool calls, rewards may come from fallible graders, and trajectories may span many partially observed steps. The classical foundation is model-based RL; this article focuses on what changes in the newer setting.
The Core Ideas
1. What each benchmark measures, when it was saturated, and what replaced it
2. Classical: Atari, MuJoCo, Procgen, Montezuma’s Revenge (links to your exploration post)
3. Reasoning: GSM8K, MATH, AIME
4. Coding: HumanEval, SWE-bench and variants
5. Agentic: WebArena, GAIA, OSWorld, τ-bench
6. Memory: MemoryAgentBench, MemoryBench
7. The framing that ties it together: a benchmark is a frozen RL environment. Same design questions, same failure modes
8. Saturation and contamination
Practical Checklist
- Define the state, action, termination condition, and reward before choosing an optimizer.
- Log raw trajectories and every reward component, not only the aggregate score.
- Separate training measurements from held-out evaluation and adversarial tests.
- Treat generated code, tool calls, and environment actions as untrusted.
- Record model, framework, benchmark, and grader versions so comparisons remain reproducible.
Limitations
This area changes quickly. Framework support and benchmark leaderboards are time-sensitive, while many recent methods are preprints rather than settled practice. The article describes mechanisms and reported evidence; it does not present an original training run.
Key Learnings
- What each benchmark measures, when it was saturated, and what replaced it
- Classical: Atari, MuJoCo, Procgen, Montezuma’s Revenge (links to your exploration post)
- Reasoning: GSM8K, MATH, AIME