Benchmark Almanac

BenchCAD 2.1

Claude Fable 5.2 83.5% leads since November 2026
In use
Version 2.1

Models reconstruct parts, assemblies, and circuit boards from drawings and images. The Core set contains 100 problems with partial credit.

Models use 60 rounds and 500,000 output tokens in the standard harness. The revised scorer makes scores incomparable with earlier versions. Core requires a Hugging Face account with access. Public samples are available.

Made by
BenchCAD team
Released
October 2026
Kind
CAD models
Level
Expert
Scored
Checked automatically
Answers
Agent task
Tasks
Public, plus a private test set
Measures
Capability
Language
English
In and out
Text, Images, 3D → 3D, Code, Actions
Interaction
Agentic
Environment
CadQuery
Run by
Anyone
Based on
BenchCAD 2.0
Strongest model tested
GPT-6.2 Astra #1 on the capability index
Last in a launch post
November 2026
Last model added
November 2026
Scores refreshed
November 20th, 2026
9
Models scored on its own leaderboard
2
Launch posts list it, from 1 lab
1
System card reports it, from 1 lab

Claude Fable 5.2 leads with 83.5%

Best score per model from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

  1. 1 Claude Fable 5.2 max 83.5%
  2. 2 GPT-6.2 Astra max 82.5%
  3. 3 GPT-6.2 Sol max 73.1%
  4. 4 Claude Opus 5.5 max 65.3%
  5. 5 GPT-6 Astra max 62.6%
  6. 6 Claude Sonnet 5.5 max 60.9%
  7. 7 GPT-6.1 Sol max 52.6%
  8. 8 Claude Fable 5.1 max 44.4%
  9. 9 Claude Haiku 5.5 max 22.9%

Source: Official BenchCAD 2.1 leaderboard

#ModelScore ± 95% CISettingSubmitted byDateSource
1 Claude Fable 5.2 83.5% max, Claude Code Anthropic Nov 2026 Official leaderboard
2 GPT-6.2 Astra 82.5% max, Codex OpenAI Nov 2026 Official leaderboard
3 GPT-6.2 Sol 73.1% max, Codex OpenAI Nov 2026 Official leaderboard
4 Claude Opus 5.5 65.3% ±1.4 max, Claude Code Anthropic Sep 2026 Official leaderboard
5 GPT-6 Astra 62.6% max, Codex OpenAI Sep 2026 Official leaderboard
6 Claude Sonnet 5.5 60.9% ±2.4 max, Claude Code Anthropic Sep 2026 Official leaderboard
7 GPT-6.1 Sol 52.6% max, Codex OpenAI Sep 2026 Official leaderboard
8 Claude Fable 5.1 44.4% ±2.2 max, Claude Code Anthropic Sep 2026 Official leaderboard
9 Claude Haiku 5.5 22.9% ±0.9 max, Claude Code Anthropic Oct 2026 Official leaderboard

Labs report scores for 3 models in their cards. We plot every source in the charts below.

The best score, 83.5%, costs $51 a task

Runs on the cost frontier score higher than any run that's cheaper. GPT-6.2 Astra at max comes within 1 points of the best for $5.40 a task, 9 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

Anthropic OpenAI Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.001 $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale Claude Fable 5.2, max GPT-6.2 Astra, max GPT-6.2 Astra, xhigh GPT-6.2 Sol, max GPT-6.2 Sol, high GPT-6.2 Sol, medium GPT-6.2 Sol, low 0% 20% 40% 60% 80% 100% $0.001 $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale Claude Fable 5.2, max GPT-6.2 Astra, xhigh

Source: Official BenchCAD 2.1 leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Fable 5.2 max Claude Code 83.5% $51
GPT-6.2 Astra max Codex 82.5% $5.40
GPT-6.2 Astra xhigh Codex 77.6% $3.40
Claude Fable 5.2 max BenchCAD 2.1 75.4% $23
GPT-6.2 Astra high Codex 74.6% $2.90
Claude Fable 5.2 xhigh Claude Code 74% $23
GPT-6.2 Sol max Codex 73.1% $1.60
GPT-6.2 Sol xhigh Codex 70.4% $1.00
Claude Fable 5.2 xhigh BenchCAD 2.1 70.2% $17
GPT-6.2 Astra medium Codex 68.5% $1.90
Claude Opus 5.5 max Claude Code 65.3% $20
GPT-6.2 Sol high Codex 64% $0.75
GPT-6 Astra max Codex 62.6% $9.70
Claude Sonnet 5.5 max Claude Code 60.9% $16
Claude Fable 5.2 high Claude Code 60.9% $10
Claude Fable 5.2 high BenchCAD 2.1 59.2% $9.00
Claude Opus 5.5 xhigh Claude Code 54.5% $9.00
GPT-6 Astra xhigh Codex 53.9% $6.30
GPT-6.1 Sol max Codex 52.6% $1.60
GPT-6 Astra high Codex 51.1% $5.40
Claude Opus 5.5 max BenchCAD 2.1 49% $8.10
Claude Sonnet 5.5 max BenchCAD 2.1 46.8% $6.10
GPT-6.1 Sol xhigh Codex 45.6% $1.10
Claude Fable 5.1 max Claude Code 44.4% $35
GPT-6.2 Astra low Codex 42.6% $0.98
Claude Opus 5.5 xhigh BenchCAD 2.1 42.4% $5.20
GPT-6.1 Sol high Codex 42.1% $0.87
GPT-6.2 Sol medium Codex 41.2% $0.29
Claude Fable 5.1 xhigh Claude Code 40.5% $28
Claude Sonnet 5.5 xhigh Claude Code 39.4% $4.00
GPT-6 Astra medium Codex 38.7% $3.70
Claude Sonnet 5.5 xhigh BenchCAD 2.1 37% $3.50
Claude Fable 5.1 max BenchCAD 2.1 36.3% $20
Claude Fable 5.1 xhigh BenchCAD 2.1 34% $17
Claude Fable 5.2 medium Claude Code 32.4% $3.00
GPT-6.1 Sol medium Codex 31.2% $0.55
Claude Opus 5.5 high Claude Code 29.7% $2.70
Claude Fable 5.2 medium BenchCAD 2.1 28.6% $2.00
Claude Fable 5.2 low Claude Code 26.8% $2.00
Claude Fable 5.1 high Claude Code 25.8% $13
Claude Opus 5.5 high BenchCAD 2.1 24.3% $2.10
GPT-6.2 Sol low Codex 23.9% $0.15
Claude Haiku 5.5 max Claude Code 22.9% $3.20
Claude Fable 5.2 low BenchCAD 2.1 21.3% $1.00
Claude Fable 5.1 high BenchCAD 2.1 20.4% $6.90
GPT-6 Astra low Codex 20.3% $1.10
Claude Opus 5.5 medium Claude Code 19.1% $1.20
Claude Haiku 5.5 max BenchCAD 2.1 18.6% $1.40
Claude Haiku 5.5 xhigh Claude Code 17.9% $1.40
Claude Sonnet 5.5 high Claude Code 16.5% $0.96
Claude Sonnet 5.5 high BenchCAD 2.1 16.1% $1.00
Claude Fable 5.1 medium Claude Code 15.7% $6.10
Claude Opus 5.5 medium BenchCAD 2.1 15.1% $0.89
Claude Haiku 5.5 xhigh BenchCAD 2.1 14.9% $0.47
GPT-6.1 Sol low Codex 13.5% $0.19
Claude Fable 5.1 medium BenchCAD 2.1 12.1% $3.20
Claude Fable 5.1 low Claude Code 12.1% $3.20
Claude Haiku 5.5 high Claude Code 11.2% $0.17
Claude Haiku 5.5 high BenchCAD 2.1 9.10% $0.06
Claude Fable 5.1 low BenchCAD 2.1 8.20% $1.50
Claude Opus 5.5 low BenchCAD 2.1 6.90% $0.15
Claude Opus 5.5 low Claude Code 6.30% $0.17
Claude Sonnet 5.5 medium BenchCAD 2.1 6.10% $0.11
Claude Haiku 5.5 medium BenchCAD 2.1 6.00% $0.01
Claude Sonnet 5.5 medium Claude Code 5.70% $0.10
Claude Haiku 5.5 medium Claude Code 5.50% $0.01
Claude Sonnet 5.5 low BenchCAD 2.1 5.20% $0.08
Claude Sonnet 5.5 low Claude Code 5.00% $0.08
Claude Haiku 5.5 low BenchCAD 2.1 4.90% $0.008
Claude Haiku 5.5 low Claude Code 4.90% $0.006

2 models have held number one on the official leaderboard since its release in October 2026

Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard and what labs claim in their cards.

Anthropic OpenAI
Own leaderboard 2 leaders
Lab claims 2 leaders

Sources: Official BenchCAD 2.1 leaderboard and labs’ model cards

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
On the benchmark’s own leaderboard Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2027 Near the ceiling (85%) Claude Fable 5.2 Claude Opus 5.5 Claude Fable 5.1 0% 20% 40% 60% 80% 100% 2027 Near the ceiling (85%) Claude Fable 5.2 Claude Opus 5.5

Sources: Official BenchCAD 2.1 leaderboard and labs’ model cards

One lab put it in a launch post in the last six months

Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out. Anthropic accounts for 7 of the 14 releases that named it since 2024.

OpenAI
DateLaunch postScoreAs the post writes it
Nov 17, 2026 GPT-6.2 Sol 73.1 BenchCAD 2.1 Core (GPT-6.2 Sol Max)
Nov 17, 2026 GPT-6.2 Astra 82.5 BenchCAD 2.1 Core (GPT-6.2 Astra Max)

One system card reports it

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

DateDocumentScoreAs the document writes itSection
Nov 9, 2026 Claude Fable 5.2 system card 75.4 BenchCAD 2.1 Core, 100 problems, no tools, adaptive thinking max effort, official scorer, five runs, up to 60 rounds and 500k output tokens. With tools 83.5%, without round or token limits. Not comparable with BenchCAD 1.0 (Section 8.13.2) Capabilities

3 lab claims next to the leaderboard

Each score a lab reported for its own model, beside the same model's best score on the leaderboard. Gaps mostly come from different harnesses, effort levels and task subsets. 2 of them match, which usually means one side copied the other. Every benchmark.

Model Lab says Board says Gap
Claude Fable 5.2 75.4% BenchCAD 2.1 Core, 100 problems, no tools, adaptive thinking max effort, official scorer, five runs, up to 60 rounds and 500k output tokens. With tools 83.5%, without round or token limits. Not comparable with BenchCAD 1.0 (Section 8.13.2) 83.5% Leaderboard, max, Claude Code −8.1 pts
GPT-6.2 Astra 82.5% BenchCAD 2.1 Core (GPT-6.2 Astra Max) 82.5% Leaderboard, max, Codex Same
GPT-6.2 Sol 73.1% BenchCAD 2.1 Core (GPT-6.2 Sol Max) 73.1% Leaderboard, max, Codex Same

Versions

We switch the top row to each new version when it comes out. Open it for one row per version. An old version stays in play while people still test it, so whoever first takes it past 85% gets credit for closing it out.

Anthropic OpenAI
BenchCAD 2.1 6 leaders
BenchCAD Passed 85% in Sep 2026, by GPT-6 Astra
GPT-6 Astra 95.9%
BenchCAD 2.0 1 leader
BenchCAD 2.1 2 leaders

Sources: Official leaderboards and labs’ model cards

More in Engineering and CAD

All 112 in Engineering and CAD
  • EEBench In use 13 electrical-engineering circuit and PCB design tasks graded by deterministic build checks, simulation and BOM constraints. No LLM judge. Gemini 4 Carbon 83.9%
  • 3DCodeBench In use Agentic procedural 3D modeling via code: author engine-ready 3D assets through software APIs. Graded for executability and shape fidelity. LongCat-2.5 55.4%
  • CADGenBench Only in system cards Agents construct CAD models from design prompts. The generation split is graded for executability and geometric correctness. Claude Fable 5 45.1%
  • PartBench Only in system cards Parametric mechanical part design and CAD-style generation, graded for geometric and manufacturing fidelity. Grok 4.6 58.5%
  • Razavi-Bench Not in model cards Models explain what happens in analog circuits and why. A language model awards up to four points per answer against a reference and a rubric. Gemini 4 Carbon 98.3%
  • CircuitRubric Not in model cards Models write analog circuit netlists from design requests. A program compares transistor connections and relative device sizes with accepted reference circuits. Claude Opus 4.8 96%

Data audit

All 12 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access Public tasks, plus a private test set.
  • Pass Confidence Checked against a primary source.
Scores 5 pass
  • Pass Independent runs Scores from the official leaderboard (9 entries), launch posts (2 scores) and system cards (1 score).
  • Pass Models scored 9 models we track.
  • Pass Newest score The newest is from November 2026, for GPT-6.2 Astra.
  • Pass Frontier models 7 of the top 10 have a score, led by GPT-6.2 Astra (#1).
  • Pass Headroom The best is 83.5%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on benchcad.com: 70 rows for models we track at the last read.
  • Pass Last read November 20th, 2026.
  • Pass Read worked The last read worked.
Consistency 2 pass
  • Pass Lab claims All 3 pairs agree within 10 points.
  • Pass Known issues No published audit found problems.

How we check