Domain on the map
AI science benchmarks and RL environments for science
Scientific research work, such as handling lab and biological data, reading the literature and reasoning about experiments.
- RL environments
- 7
- AI benchmarks
- 70
- Shares entries with
- 18 categories
Snapshot 2026-11-05. How categories are built · Where the data comes from
In short
- What are RL environments for science?
- A simulated workplace where an AI agent practises a task (a code repository, a browser, business software) and an automatic grader scores each attempt. Labs train on thousands of these attempts. Bought to improve models. For science, that means scientific research work, such as handling lab and biological data, reading the literature and reasoning about experiments.
- Who builds RL environments for science?
- Surge AI, Latch, Emulated, ReasonCore, Ulam and Rise Data Labs, among 7 confirmed below.
- What are the main science benchmarks for LLMs?
- ProteinGym, BioMysteryBench Human-Difficult, BioMysteryBench Human-Solvable, BixBench, FormationEval and SWE, among 65 confirmed below.
RL environments 7
-
Surge AI
Sells RL environments“We create rich, complex RL environments that challenge agentic models in novel ways, and design verifiers that reward their behavior.” source ↗
-
Latch
No line from its own site or paper places it in Science yet; the catalog lists it here.
- Emulated
- ReasonCore
- Ulam
-
Rise Data Labs
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
Tacit Labs
Covers science“Tacit Labs is building data, evaluations, and research environments unlocking scientific discovery capabilities for frontier models.” source ↗
Sells RL environments“help build Tacit's scientific agent environments: integrated environments where language-model agents can use tools, inspect data, make decisions” source ↗
AI benchmarks 70
-
ProteinGym
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
BioMysteryBench Human-Difficult
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
BioMysteryBench Human-Solvable
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
BixBench
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
FormationEval
Tests science“an open multiple-choice question benchmark for evaluating language models on petroleum geoscience and subsurface disciplines.” source ↗
-
SWE
Emulated · built it“We're introducing the **Software Development Automation Benchmark**: 80 tasks that grade not just code generation” source ↗
-
Terminal-Bench-Science
Tests science“A benchmark for evaluating AI agents on research workflows across scientific domains” source ↗
-
Math RL Tasks
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
ChemBench
Tests science“ChemBench is a comprehensive benchmark for evaluating language and multimodal models on chemical reasoning, knowledge, and intuition” source ↗
-
scBench
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
SciKnowEval
Tests science“We have recently added the Physics and Materials to SciKnowEval.” source ↗
-
SpatialBench
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
TDC AMES
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
TDC BBB_Martins
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
TDC Bioavailability_Ma
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
TDC Caco2_Wang
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
TDC Clearance_Hepatocyte_AZ
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
TDC Clearance_Microsome_AZ
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
TDC CYP2C9_Substrate_CarbonMangels
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
TDC CYP2C9_Veith
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
TDC CYP2D6_Substrate_CarbonMangels
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
TDC CYP2D6_Veith
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
TDC CYP3A4_Substrate_CarbonMangels
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
TDC CYP3A4_Veith
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
TDC DILI
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
TDC DRD3 Docking
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
TDC Half_Life_Obach
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
TDC hERG
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
TDC HIA_Hou
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
TDC LD50_Zhu
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
TDC Lipophilicity_AstraZeneca
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
TDC Pgp_Broccatelli
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
TDC PPBR_AZ
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
TDC Solubility_AqSolDB
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
TDC VDss_Lombardo
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
VariantBench
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
GeneBench
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
PhysicsFinals
Tests science“We introduce PHYSICS, a comprehensive benchmark for university-level physics problem solving.” source ↗
-
PROBE Binding Affinity
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
PROBE Function Prediction
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
PhysicsMind
Tests science“a unified benchmark with both real and simulation environments that evaluates law-consistent reasoning and generation over three canonical principles: Center of Mass, Lever Equilibrium, and Newton's First Law.” source ↗
-
A^3-Bench
Tests science“a benchmark designed to evaluate scientific reasoning through dual-scale memory-driven activation, grounded in Anchor and Attractor Activation.” source ↗
-
ActiveSciBench
Tests science“ActiveSciBench-Chem with 57 enzyme-kinetics tasks” source ↗
-
AgroBench
Tests science“a benchmark for evaluating VLM models across seven agricultural topics, covering key areas in agricultural engineering and relevant to real-world farming.” source ↗
-
FrontierChallenge
Tests science“spanning quantum chemistry, molecular dynamics, materials characterization, analytical chemistry, life science, and electrochemistry/environment” source ↗
-
SCICONVBENCH
Tests science“a benchmark for multi- turn clarification in scientific task formulation across four computational science problem domains: fluid mechanics, solid mechanics, materials science” source ↗
-
TDC BindingDB Patent DTI-DG
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
TDC TCHard TCR-Epitope
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
CUSP
Tests science“a multi-disciplinary and event-level benchmark that evaluates scientific forecasting in AI systems through feasibility assessment, mechanistic reasoning, generative solution design, and temporal prediction.” source ↗
-
Geo-Eval
Tests science“Our extensive evaluation on a novel domain-specific benchmark, Geo-Eval, reveals that a domain-aligned 8B model can outperform open-weight 70B generalists” source ↗
-
GSO-Bench
Tests science“General science and observation benchmark” source ↗
-
gwBenchmarks
Tests science“We introduce gwBenchmarks, a suite of eight tasks grounded in gravitational wave analytic calculations and numerical simulations” source ↗
-
MaCBench
Tests science“MaCBench is a benchmark for multimodal language models in chemistry and materials research.” source ↗
-
MLReplicate
Tests science“we introduce MLReplicate, an end-to-end benchmark evaluating autonomous research systems on machine learning reproducibility.” source ↗
-
NatureBench
Tests science“a cross-discipline benchmark of 90 tasks distilled from peer-reviewed Nature-family publications” source ↗
-
Proof of Time
Tests science“we introduce PoT, a semi-verifiable benchmarking framework that links scientific idea judgments to downstream signals that become observable later” source ↗
-
SoundnessBench
Tests science“a curated benchmark of 1,099 machine-learning research proposals reconstructed from ICLR submissions, labeled with reviewer soundness sub-scores, and audited against source papers.” source ↗
-
LAB-Bench
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
PDEBench
Tests science“We introduce PDEBench, a benchmark suite of time-dependent simulation tasks based on Partial Differential Equations (PDEs).” source ↗
-
Collider-Bench
Tests science“a benchmark for evaluating whether LLM agents can reproduce experimental analyses from the Large Hadron Collider (LHC) using only public papers and open scientific software” source ↗
-
EngiBench
Tests science“We introduce EngiBench, a hierarchical benchmark designed to evaluate LLMs on solving engineering problems.” source ↗
-
MoleculeNet
Tests science“For quantum mechanical and biophysical datasets, the use of physics-aware featurizations can be more important than choice of particular learning algorithm.” source ↗
-
OpenADMET ExpansionRx
No line from its own site or paper places it in Science yet; the catalog lists it here.
-
Prediction Bottlenecks Don't Discover Causal Structure
Tests science“standardized synthetic generators (VAR/Lorenz/CauseMe-style), three intervention semantics (do(X=c), soft-noise, random-forcing), edge-provenance cards on three real datasets” source ↗
-
SciBench
Tests science“SciBench contains a carefully curated dataset featuring a range of collegiate-level scientific problems from mathematics, chemistry, and physics domains.” source ↗
5 not yet confirmedListed in the catalog; no line from its own site or paper confirms it yet.
Data 14
- Hillclimb
- Abundant
- Aviro
- Cua
-
Emulated
Sells data“We provide ready-to-use datasets across software engineering, machine learning engineering, autonomous research, and model post-training.” source ↗
Covers science“ready-to-use datasets across software engineering, machine learning engineering, autonomous research, and model post-training” source ↗
- Mercor
- Pareto
- Patronus AI
- ReasonCore
- Revelo
- Symbal
- Turing
- Ulam
- Vetto AI
Next: pick another domain. Tags come from the RL Research daily scrape, canonicalized nightly.