CFDLLMBench
Three-part benchmark for LLMs in CFD: graduate-level CFD knowledge questions, numerical reasoning in code, and context-dependent OpenFOAM workflow implementation, scored on code executability, solution accuracy, and numerical convergence.
What it does and how it connects
CFDLLMBench is a three-part benchmark for large language models in computational fluid dynamics. CFDQuery tests conceptual understanding with 90 graduate-level multiple-choice questions, CFDCodeBench tests instruction following with 24 natural-language prompts for CFD code, and FoamBench tests end-to-end OpenFOAM automation from natural language with 110 basic and 16 advanced simulation tasks.
Scoring covers code executability, solution accuracy and numerical convergence, and the V2 datasets ship in the repository, with the original paper datasets also on Kaggle. The benchmark is BSD-3-Clause licensed, and the paper is published in the Journal of Data-centric Machine Learning Research, volume 3, number 13.
FoamBench benchmarks the full chain from case creation through execution to postprocessing, assessing multi-step reasoning, domain adaptation and recovery. The V2 update refreshed prompts for some cases, evaluated newer LLMs and evaluated the latest version of Foam-Agent.
Alternatives
Compare side by side- P3D-Bench
A CAD benchmark scored on geometry and part coverage, rather than a benchmark for LLMs in computational fluid dynamics.
- StepScore CAD Benchmark
A 44-metric CAD benchmark scored against human-calibrated thresholds, rather than a benchmark for LLMs in CFD.
- Easy-CAD / bendcad (jnadeau207-collab)
An agent-native CAD kernel, rather than a benchmark for LLMs in computational fluid dynamics.
- PartForge evals
An eval suite for verification workflows, rather than a benchmark for LLMs in CFD.
- Rapidflare suite
Rapidflare's internal eval practice for AI agents in electronics, rather than a benchmark for LLMs in computational fluid dynamics.
Sources
Not verified
- Pricing: none stated on the repo or project page.
- Real user voices: no real linked post read for this page; the repo's only issue is a Hugging Face outreach ask, not a user voice.
- No GitHub releases exist, so latest release fields are null.
- No OpenFOAM error ids exist in the error map, so no related errors are recorded.
- The README links the contributing guide to an organization spelled NLR-Theseus, which looks like a typo for NREL-Theseus; the repo URL with NREL-Theseus is the one that works.
Last verified November 6, 2026. Listings are informational. The evidence score measures public, verifiable evidence and is not an endorsement or a compatibility claim.