Catalog Systems Memory benchmarks & evaluation NetHack Learning Environment (NLE)

NetHack Learning Environment (NLE)

https://github.com/heiner/nle

DeepMind / FAIR's NetHack Learning Environment — a Gym-compatible interface to NetHack, the canonical roguelike. Massive procedural state space, permadeath, deep item / skill interactions, ASCII partial-observability. Released at NeurIPS 2020; canonical long-horizon partial-observability RL benchmark.

Active · Verified 2026-05-14 · fresh as of 2026-05-14 · 31 cited facts, 0 with a third-party source · Suggest a correction

At a glance

Type
Roguelike RL benchmark — NetHack as agent environment repo fix
Tier
T2
Section
Memory benchmarks & evaluation
Created
2020-06 (Küttler et al. NeurIPS 2020) repo fix
Latest release
v0.9.x series (active as of 2024) repo fix
License
NetHack General Public License repo fix
GitHub
1.0k★ Python/C repo fix
Pricing
not applicable — open benchmark; free to use repo fix
Funding
not applicable — not commercial fix
Freshness
2026-05-14 fix
Layer
agent-infra repo fix
Maintainer health
1 issues in 90 days · 1 release in 12 months (latest 2026-04-22) · bus factor 2 — as of 2026-11-01 repo

Memory taxonomy

storage
n/a
retrieval
n/a
persistence
n/a
update
read-only
unit
n/a
governance
n/a
conflict
n/a

When to use

Optimised for: not applicable - eval dataset, not a system

Anti-fit: not applicable - eval dataset, not a system

Pros & cons

Pros

Massive (10^48+) state space + long horizon (10^4-10^5 steps) makes it one of the hardest RL benchmarks; ASCII representation is interpretable; procedural so memorisation cannot solve it.

Cons

Sparse reward + dense procedural complexity → RL still hard-stuck at sub-human levels; ASCII-only modality limits transfer to multimodal agents.

Claims & capabilities

NeurIPS 2020 paper; NetHack Challenge run at NeurIPS 2021. After 4 years still considered unsolved at expert-human level by autonomous agents.

Technical surface

API surface
not applicable — eval dataset, not a system
Backend storage
not applicable — eval dataset, not a system
Deployment
not applicable — run locally
Embedding model
not applicable — eval dataset, not a system
Multi-tenancy
not applicable — eval dataset, not a system
MCP
not applicable — benchmark / evaluation harness
A2A
not applicable — benchmark / evaluation harness
OpenTelemetry
not applicable — benchmark / evaluation harness

Similar systems

Other memory benchmarks & evaluation in the catalog, ranked by inbound references.

  • LoCoMo T3

    Conversations spanning up to 35 sessions / 300 turns / ~9K tokens, built from LLM-agent personas grounded on temporal event graphs and verified by human annotators. Tests QA (single-hop, multi-hop, temporal, commonsense, adversarial), event-graph summarisation, multimodal dialogue.

  • GAIA T3

    Benchmark for general AI assistants made of real-world questions that require reasoning, multimodality, web browsing and tool use, with a public leaderboard on Hugging Face. Traction: ~917 likes on the Hugging Face dataset

  • LongBench T3

    Tsinghua THUDM. 21 datasets across 6 categories (single-doc QA, multi-doc QA, summarisation, few-shot ICL, synthetic, code completion) in English + Chinese. ~6.7K words English / 13.4K chars Chinese average. v2 at ACL 2025 with harder, longer instances.

  • LongMemEval T3

    5 core memory abilities — info extraction, multi-session reasoning, temporal reasoning, knowledge updates, abstention — via 500 curated questions in synthetic chat histories. Two scales: LongMemEval-S (115K tokens), LongMemEval-M (up to 1.5M).

  • RULER T3

    NVIDIA. Extends NIAH with multi-needle retrieval, multi-hop entity tracing, aggregation, QA — configurable for arbitrary length / difficulty. Evaluated 17 long-context models on 13 tasks.

  • InfiniteBench (∞Bench) T3

    Tsinghua / OpenBMB. First benchmark with average input >100K tokens. 12 tasks across English + Chinese: long novel QA, code debugging, math in long context, summarisation, retrieval. Realistic documents (novels, code) rather than synthetic filler.

Sources (30 cited facts, 0 third-party)

Each fact links to the page it came from; the badge says who published it (how to read sources).