BenchCAD
the entrance to AI for hardware

A Benchmark for Programmatic CAD

A benchmark for code-based 3D modeling on professional tasks.

read the 2.1 post run the benchmark Core access
BenchCAD T1 · drawing → part
impeller, a BenchCAD 1.0 part family, turning coil spring, a BenchCAD 1.0 part family, turning double simplex sprocket, a BenchCAD 1.0 part family, turning spline hub, a BenchCAD 1.0 part family, turning handwheel, a BenchCAD 1.0 part family, turning
taskT1
parts1
chosen by:
Score · Vision2Code

Where the frontier is

Vision2Code · four rendered views in, a CadQuery program out
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 $0.001 $0.003 $0.01 $0.03 $0.1 $0.3 $1 $3 $10 cost per task, USD (log scale) score Claude Sonnet 5 · with tools · 0.199 at $0.0042 to 0.518 at $6.8838 per task — read off Anthropic's Opus 5.5 system-card fig 8.13.2.B Claude Opus 5 · with tools · 0.303 at $0.0178 to 0.898 at $10.0777 per task — read off Anthropic's Opus 5.5 system-card fig 8.13.2.B Claude Fable 5.1 · with tools · 0.383 at $0.0251 to 0.926 at $10.4927 per task — read off Anthropic's Opus 5.5 system-card fig 8.13.2.B Claude Opus 5.5 · with tools · 0.397 at $0.0092 to 0.961 at $5.937 per task — read off Anthropic's Opus 5.5 system-card fig 8.13.2.B Claude Sonnet 5.5 · with tools · 0.330 at $0.0039 to 0.963 at $4.71 per task — read off Anthropic's Haiku 5.5 system card fig. 8.9.2.B (updated cache-read price, 1,000-file subset) GPT-6 Astra · with tools · 0.672 at $0.05 to 0.959 at $1.75 per task — digitised from OpenAI's BenchCAD launch figure GPT-5.6 Sol · with tools · 0.360 at $0.05 to 0.834 at $2.9 per task — digitised from OpenAI's BenchCAD launch figure Grok 4.5 · with tools · 0.446 at $0.14 to 0.777 at $4.96 per task — our own run on our harness Claude Haiku 5.5 · with tools · 0.323 at $0.0007 to 0.870 at $1.63 per task — read off Anthropic's Haiku 5.5 system card fig. 8.9.2.B (1,000-file subset, list prices) Claude Haiku 4.5 · with tools · about 0.21 at $0.11–0.13 per task, fixed extended-thinking budgets up to 32k — read off Anthropic's Haiku 5.5 system card fig. 8.9.2.B Pareto frontier: the best score reached at or below each cost, across all models on this chart Pareto frontier: best score at each cost Claude Sonnet 5 · $0.004 per task · score 0.20 Claude Sonnet 5 · $0.21 per task · score 0.24 Claude Sonnet 5 · $1.33 per task · score 0.38 Claude Sonnet 5 · $2.70 per task · score 0.45 Claude Sonnet 5 · $6.88 per task · score 0.52 Claude Opus 5 · $0.018 per task · score 0.30 Claude Opus 5 · $0.74 per task · score 0.47 Claude Opus 5 · $3.56 per task · score 0.77 Claude Opus 5 · $6.82 per task · score 0.87 Claude Opus 5 · $10.1 per task · score 0.90 Claude Fable 5.1 · $0.025 per task · score 0.38 Claude Fable 5.1 · $0.11 per task · score 0.45 Claude Fable 5.1 · $0.58 per task · score 0.55 Claude Fable 5.1 · $5.17 per task · score 0.83 Claude Fable 5.1 · $10.5 per task · score 0.93 Claude Opus 5.5 · $0.009 per task · score 0.40 Claude Opus 5.5 · $0.025 per task · score 0.53 Claude Opus 5.5 · $0.059 per task · score 0.56 Claude Opus 5.5 · $0.44 per task · score 0.76 Claude Opus 5.5 · $5.93 per task · score 0.96 Claude Sonnet 5.5 · $0.004 per task · score 0.33 Claude Sonnet 5.5 · $0.004 per task · score 0.34 Claude Sonnet 5.5 · $0.011 per task · score 0.47 Claude Sonnet 5.5 · $0.16 per task · score 0.70 Claude Sonnet 5.5 · $4.71 per task · score 0.96 GPT-6 Astra · $0.050 per task · score 0.67 GPT-6 Astra · $0.35 per task · score 0.90 GPT-6 Astra · $0.85 per task · score 0.94 GPT-6 Astra · $1.75 per task · score 0.96 GPT-5.6 Sol · $0.050 per task · score 0.36 GPT-5.6 Sol · $0.20 per task · score 0.48 GPT-5.6 Sol · $0.50 per task · score 0.64 GPT-5.6 Sol · $1.20 per task · score 0.75 GPT-5.6 Sol · $2.90 per task · score 0.83 Grok 4.5 · $0.14 per task · score 0.45 Grok 4.5 · $0.92 per task · score 0.55 Grok 4.5 · $4.96 per task · score 0.78 Claude Haiku 5.5 · $0.001 per task · score 0.32 Claude Haiku 5.5 · $0.001 per task · score 0.35 Claude Haiku 5.5 · $0.008 per task · score 0.49 Claude Haiku 5.5 · $0.25 per task · score 0.79 Claude Haiku 5.5 · $1.63 per task · score 0.87 Claude Haiku 4.5 · $0.11 per task · score 0.20 Claude Haiku 4.5 · $0.11 per task · score 0.21 Claude Haiku 4.5 · $0.12 per task · score 0.21 Claude Haiku 4.5 · $0.13 per task · score 0.21 Sonnet 5.5 Opus 5.5 GPT-6 Astra Fable 5.1 Opus 5 GPT-5.6 Sol Grok 4.6 Grok 4.5 Sonnet 5 Haiku 5.5 Haiku 4.5 Grok 4.6 · with tools · 0.805 at $6.979/task — our own run on our harness lowmediumhighmediummediumxhighxhighhighxhighmaxmaxHaiku 5.5Opus 5.5Sonnet 5.5GPT-6 Astra
Leaderboard

The top of the board

The ten best models on Vision2Code. Scores run from 0 to 1, and a program that fails to run scores 0.


All four tasks
    * Reported by frontier labs
    Latest

    News and writing

    November 3, 2026 · News

    OpenAI on BenchCAD 2.0: a newer internal model scores 0.875
    October 23, 2026 · News

    Kimi K3.1 Pro on BenchCAD: 0.927 with tools, 0.728 without
    October 12, 2026 · Blog

    Introducing BenchCAD 2.0

    Citation

    @misc{benchcad2_2026,
  title        = {BenchCAD 2.0: Benchmarking Programmatic CAD on Real Engineering Work},
  author       = {{BenchCAD Team}},
  year         = {2026},
  howpublished = {\url{https://benchcad.com/blog/benchcad-2.html}},
  note         = {Blog post}
}