Engineering and CAD
We track 26 benchmarks in engineering and CAD. They run from university level to research frontier level.
OpenAI held the top engineering and CAD score 62% of the contested time
That's about 29 of the 46 months since January 2023, averaged over these benchmarks. We only count time while other labs are still testing, and stop once a model passes 85%. Faded stretches no longer count.
Google DeepMind OpenAI Anthropic
University Undergraduates and entry-level professionals
Expert Specialists with years of training, often over hours of work
Research frontier Researchers, on open problems or tasks that take days
Sources: Official leaderboards and labs’ model cards
Share of contested time at number one
OpenAI 62% Anthropic 31% 
Google DeepMind 7%
Google DeepMind 7% Sources: Official leaderboards, Epoch AI (CC BY 4.0) and labs’ model cards
Each benchmark counts from the one source its own page leads with: its official leaderboard when that is current, else Epoch AI’s own runs, else labs’ own numbers. We date independent runs by the model’s release and lab reports by when they came out.
Google DeepMind OpenAI Anthropic
University Undergraduates and entry-level professionals
Expert Specialists with years of training, often over hours of work
Research frontier Researchers, on open problems or tasks that take days
Source: Official leaderboards
Share of contested time at number one, independent runs only
OpenAI 58% Anthropic 34% Google DeepMind 8%
Google DeepMind 8% Sources: Official leaderboards and Epoch AI (CC BY 4.0)
Every independent run: each benchmark’s own leaderboard, Epoch AI’s runs and boards Epoch AI copied, dated by the model’s release. We leave out benchmarks that only labs report.
Google DeepMind OpenAI Anthropic
University Undergraduates and entry-level professionals
Expert Specialists with years of training, often over hours of work
Research frontier Researchers, on open problems or tasks that take days
Sources: Official leaderboards and labs’ model cards
Share of contested time at number one, claimed scores included
OpenAI 52% Anthropic 42% Google DeepMind 6%
Google DeepMind 6% Sources: Official leaderboards, Epoch AI (CC BY 4.0) and labs’ model cards
Every score, including labs’ own numbers and claims Epoch AI collected from reports and papers. Labs pick their own settings, so a claim can beat an independent run of the same model.
From university to research frontier level
Grouped by who could solve the tasks, from things most adults can do to open research problems. The bar under each name is its best score.
Expert
Specialists with years of training, often over hours of work
Another 1 has no level. Levels don’t apply to arenas, safety evaluations or composite indices.
By kind
CAD models
| Benchmark | Level | Scored | Best score | Status |
|---|---|---|---|---|
| LMBuild Language model agents design 3D structures as parts, joints, materials and assembly steps. Scoring checks that the structures can be built and work as intended. | Expert | Checked automatically | 66.5% | Not in cards |
| Physical-Engineering's Last Exam Agents produce billable engineering deliverables such as calculation packages, drawings and design reviews in 85 tasks across electrical, civil, structural and mechanical engineering. Automatic graders check constraints, safety limits and the design objective. | Expert | Checked automatically | 0.49 | Not in cards |
| BenchCAD Programmatic CAD: 17,900 execution-verified CadQuery programs over 106 industrial part families. Four matched tasks scored on geometry. | University | Checked automatically | 95.9% | In use |
| RWE-bench Agents do mechanical, electrical and aerospace design tasks, such as a crankshaft, dense PCB routing or turbine-vane cooling, with CAD and CAE tools. A deterministic verifier runs physics solvers and passes a design only if it meets every requirement. | Expert | Checked automatically | 62% | Not in cards |
| 3DCodeArena Models generate 3D assets through code. Human preference votes determine the published Bradley-Terry Elo rating. | Expert | Ranked by votes | 1,230 | Not in cards |
| BlenderWorld Visual agents turn a text, image or video brief into an editable Blender scene with rendered media. Hidden binary rubrics and eight visual quality dimensions score each project. | Expert | Graded by a model | 79.9 | Not in cards |
| CadEval Models write OpenSCAD code for single mechanical parts from text descriptions. Rendered parts are checked against reference geometry, scored by pass rate. | University | Checked automatically | 74% | Not in cards |
| CADGenBench Agents construct CAD models from design prompts. The generation split is graded for executability and geometric correctness. | University | Checked automatically | 45.1% | System cards only |
| CAD Arena Agents rebuild licensed engineering drawings as editable CAD parts through the scripting APIs of NX, SolidWorks, Onshape, Fusion or Build123d. A verifier scores geometric match. | Expert | Checked automatically | 0.75 | Not in cards |
| Parametric CAD Bench Agents author editable parametric FreeCAD part designs, graded against held-back references for geometry and spec consistency. | Expert | Checked automatically | 61% | In use |
| EngiWorld Agents work through the full industrial engineering design loop on 1,301 expert-curated tasks that need reasoning over geometric and physical constraints across software. | Expert | Checked automatically | 44.7% | Not in cards |
| FrontierCAD Assemblies Agents place supplied CAD parts into 16 assemblies. The score is the published mean assembly geometry score. | Expert | Checked automatically | 0.71 | Not in cards |
| FrontierCAD Parts Agents build 39 CAD parts from engineering drawings. The score is the published mean geometry score. | Expert | Checked automatically | 0.63 | Not in cards |
| VWE-Bench Multimodal agents build interactive 3D worlds from a user request. They infer intent, plan the scene layout, call 3D tools for asset retrieval, editing and rendering, and revise from rendered feedback. The benchmark has 2,616 3D assets, 323 human-annotated seed worlds and 6,828 queries. Verified queries have ground truth, and the rest are scored with rubrics. | Expert | Graded by a model | 57.3% | Not in cards |
| BenchCAD 2.0 Models reconstruct parts, assemblies, and circuit boards from drawings and images. The Core set contains 100 problems with partial credit. | Expert | Checked automatically | 59% | Not in cards |
| BenchCAD 2.1 Models reconstruct parts, assemblies, and circuit boards from drawings and images. The Core set contains 100 problems with partial credit. | Expert | Checked automatically | 83.5% | System cards only |
| 3DCodeBench Agentic procedural 3D modeling via code: author engine-ready 3D assets through software APIs. Graded for executability and shape fidelity. | Ranked by votes |  55.4% | In use | |
| PartBench Parametric mechanical part design and CAD-style generation, graded for geometric and manufacturing fidelity. | University | Checked automatically | 58.5% | System cards only |
| BenchCAD 2.1-beta Models reconstruct parts, assemblies, and circuit boards from drawings and images. The Core set contains 100 problems with partial credit. | Expert | Checked automatically | Not in cards | |
| FrontierCAD Agents build 39 machined parts from engineering drawings and place parts into 16 assemblies, in their own coding harnesses. A verifier that tolerates rotation and checks hole positions scores the geometry. | Expert | Checked automatically | Not in cards | |
| SimRepairBench Agents find and repair modeling errors in 3D OpenUSD assets for robot simulation, from 25 damaged objects in NVIDIA's SimReady collection, with partial credit from verifiers. | Expert | Checked automatically | Not in cards |
Electronics
| Benchmark | Level | Scored | Best score | Status |
|---|---|---|---|---|
| EEBench 13 electrical-engineering circuit and PCB design tasks graded by deterministic build checks, simulation and BOM constraints. No LLM judge. | Expert | Checked automatically | 83.9% | In use |
| HW-Bench Agents complete six hardware engineering tasks for a 2.45 GHz antenna using native design and simulation tools. Code checks and calibrated model judges grade outputs against engineer-written criteria. The headline pass rate measures runs that satisfy every critical criterion, averaged across tasks. | Expert | Graded by a model | 67% | Not in cards |
| Build-A-Chip-Bench Agents take a chip from workload specification to a routed GDS layout with open-source tools. Each design must pass hidden verification, routability and timing gates. | Research frontier | Checked automatically | 16.6% | Not in cards |
| SimuVerity Agents build executable Simulink models from text for 101 tasks in ten engineering domains. Simulations check that each model meets its engineering requirements. | Expert | Checked automatically | 42.9 | Not in cards |
| CVDP Models and agents solve hardware design problems in Verilog, such as spec to RTL, code completion, debugging and testbench writing. Simulation tests check the outputs. | Expert | Checked automatically | Not in cards |