AIEvals

Benchmarks

BenchCAD Vision2Code (with tools)

Generating executable CadQuery code from multi-view renders with tools, measured by voxel IoU on a 1,000-file subset.

CAD reasoningReported by AnthropicVendor-reported onlyReported by Anthropic

How it is run

Section 8.9.2 gives each model a container, image files, installed libraries and an image-cropping tool; five max-effort runs are averaged on the 0–1 IoU scale.

What it cannot tell you

This tool-assisted subset is separate from no-tools Vision2Code and the other three BenchCAD tasks. IoU is not percent-correct accuracy.

Why there is no independent table

No independent suite in this dataset runs BenchCAD Vision2Code (with tools). The numbers below are the ones model publishers report about their own models, which is a different kind of evidence: each publisher chooses its own harness and effort level, so a row is readable on its own and not against the other rows.

Read this with

Other versions of the same evaluation. Versions are not comparable: the task set changes between them.

  • BenchCAD Vision2Code (no tools)Same family. Generating executable CadQuery code from multi-view renders, scored by voxel intersection over union on a 1,000-file subset.

As reported by each publisher

Kept separate per publisher, because the settings are not shared.

Anthropic

Claude Haiku 5.5: system card capability evaluations · 7 October 2026

RankModelScore (score)
3Claude Haiku 5.50.87
5Claude Haiku 4.50.21
1Claude Sonnet 5.50.963
2Claude Opus 5.50.962
4Claude Sonnet 50.519

Ranks compare reported scores in the same published row; ties share a rank.

BenchCAD Vision2Code (with tools). §8.9.2 / Figure 8.9.2.A: container and image-cropping tool, same 1,000-file subset, five max-effort runs, voxel IoU.

Section 8, pages 111–139 of the Haiku 5.5 system card, read 8 October 2026. Values come from exact prose, tables and labelled figures. Unless specified, Haiku uses adaptive thinking at max effort and five trials with evaluation-specific context up to 1M. Each row retains its subset, metric, effort and source notes; no score is assigned to an unspecified SWE-Bench Pro version. Terminal runs use safeguards without fallback and no internet egress. ProgramBench uses filtered tasks and hidden-test pass rate without a time limit. OSWorld partial and strict results use the official 82-task offline subset. CAD uses 0–1 voxel IoU. Healthcare raw and length-adjusted scores stay separate. Life-science results disable Haiku biology safeguards. Independently run figures quoted by Anthropic remain publisher reports here. Missing comparison values are not inferred from rounded charts.

www.anthropic.com/claude-haiku-5-5-system-card

Sources

Every page used for this benchmark, read on 17 October 2026.