STEM datasets for post-training evaluation and reasoning
Human-authored datasets, benchmarks, and tools for evaluating and improving scientific and mathematical reasoning in LLMs.
STEM datasets
Human-authored datasets across STEM domains to support scientific accuracy, alignment training, and symbolic rigor at scale.
Math, Physics, Chemistry, and Biology Datasets
Datasets curated to test logical structure, problem-solving accuracy, and formal rigor, grounded in real-world scientific domains.
Request STEM Data Packs
Chain-of-Thought + Stepwise Reasoning Packs
Trace-based reasoning examples scored for fidelity, designed for training and reward shaping.
Request CoT Datasets
High-Throughput Training Data
Fast, structured datasets optimized for SFT, RLHF, and symbolic alignment workflows.
Request Sample Datasets
Lean-Based Proof QA Datasets
Iterative proof generation in Lean 4 paired with informal math questions, supporting symbolic reasoning and fine-tuned verification.
Request Symbolic Reasoning DatasetsBenchmarks and evaluation
Rubric-aligned benchmarks and structured diagnostics that surface STEM-specific model weaknesses and reasoning gaps.
VLM-Bench
Benchmark model reasoning on over 700 vision–language tasks grounded in STEM, logic, and world knowledge.
Download Report
GPQA, AIME, and MMLU-Pro Comparisons
See how your models stack up on known benchmarks or define your own test sets with domain-specific metrics.
Run a Diagnostic
High-Difficulty STEM Benchmarks
Evaluate your model’s capability on problems unsolvable by SOTA models, paired with rubric-based grading and expert-written answers.
Run a DiagnosticRL environments for STEM workflows
Evaluate agents on real-world STEM tasks, generate fine-tuning trajectories, and train reward models in reproducible, high-fidelity environments.
UI-Based RL Environments for Interface Agents
Evaluate scientific reasoning agents within virtual lab environments that simulate physics, chemistry, or biological systems.
Request UI Agent Environments
MCP Environments for Function-Calling Agents
Train agents on function calling and tool execution inside sandboxed server environments. Includes tool schemas, reward verifiers, and seed databases.
Request Function-Calling EnvironmentsEnd-to-End Evaluation and Training Loops
Each RL environment includes prompts, verifiers, analytics harnesses, and trajectory outputs, enabling evaluation diagnostics, reward shaping, and supervised fine-tuning at scale.
Request RL EnvironmentsResearch and case studies
Scale STEM reasoning with expert-built datasets
Train, fine-tune, or evaluate models on structured STEM tasks, backed by domain-reviewed data and traceable QA.