COSMO-Agent
Industry-aligned CAD-CAE benchmark of executable parametric design tasks across 25 component categories; the agent revises geometry under physics, geometry, and cost constraints within fixed tool-call and retry budgets, scored on feasibility, iteration efficiency, and failure recovery.
What it does and how it connects
COSMO-Agent is a tool-augmented reinforcement learning framework for closed-loop CAD-CAE optimization: it casts CAD generation, CAE solving, result parsing and geometry revision as an interactive RL environment, where a language model learns to orchestrate external tools and revise parametric geometries until constraints are satisfied.
A multi-constraint reward jointly encourages feasibility, toolchain robustness and structured output validity, and the paper contributes an industry-aligned dataset of about 20,000 executable CAD-CAE tasks across 25 component categories, with standardized interfaces and fixed tool-call and retry budgets. Experiments report that training substantially improves small open-source LLMs for constraint-driven design, exceeding large open-source and strong closed-source models in feasibility, efficiency and stability.
The catalog record describes the benchmark contribution; the paper itself presents COSMO-Agent as a training framework plus the benchmark, not a benchmark alone. No public code repository was found for this paper at research time, so no repo, repo stats or license are recorded for this page.
Alternatives
Compare side by side- CADEngBench
A two-track benchmark of parametric parts and assemblies scored on B-Rep validity and matched FEA, rather than a tool-augmented RL framework with a CAD-CAE task benchmark.
- Engine Bench 0.1
Two models race head to head to build one engine from one prompt, rather than a closed-loop CAD-CAE framework with about 20,000 tasks.
- cad-silent-failure-bench
Measures silent failures where CAD code runs but the part is wrong, rather than a closed-loop CAD-CAE framework.
- P3D-Bench
A CAD benchmark scored on geometry and part coverage, rather than a closed-loop CAD-CAE framework.
- StepScore CAD Benchmark
A 44-metric CAD benchmark scored against human-calibrated thresholds, rather than a closed-loop CAD-CAE framework.
Sources
Not verified
- Pricing: none stated on the paper page.
- Real user voices: no real linked post read for this page; this is a paper with no code repository found.
- License: no license is recorded; the catalog license of source-available is unverified because no public code repository was found.
- No public repo was found, so no repo stats are recorded.
Last verified November 6, 2026. Listings are informational. The evidence score measures public, verifiable evidence and is not an endorsement or a compatibility claim.