Benchmark Almanac

BIM-Edit

Gemini 4 Argon 87.7% leads since September 2026
Not in model cards

Agents create, change, or delete elements in IFC building models. The evaluator checks the resulting geometry, attributes, and relationships.

The authors define 324 edits across 11 realistic models and 36 synthetic scenes. ONESTRUCTION used its own grader and raised the turn limit.

Released
June 2026
Kind
CAD models
Level
University
Scored
Checked automatically
Answers
Agent task
Tasks
All public
Measures
Capability
Language
English
In and out
Text, 3D → Actions, 3D
Interaction
Agentic
Environment
APIs and tools
Refresh
One fixed set
Run by
Anyone
Strongest model tested
GPT-6 Astra #4 on the capability index
Last model added
October 2026
Scores refreshed
November 20th, 2026
20
Models scored on ONESTRUCTION’s leaderboard

Gemini 4 Argon leads with 87.7%

Best score per model from ONESTRUCTION’s 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 Gemini 4 Argon medium 87.7%
  2. 2 GPT-6 Astra medium 84.3%
  3. 3 DeepSeek V4.1 Flash 1017 84.3%
  4. 4 GPT-6.1 Sol medium 83.2%
  5. 5 Claude Opus 5 medium 81.9%
  6. 6 Claude Opus 5.5 medium 81.6%
  7. 7 Muse Spark 1.3 medium 81.4%
  8. 8 Claude Fable 5 medium 79.7%
  9. 9 GLM-5.3-Flash 78.9%
  10. 10 Gemini 3.7 Flash medium 77.6%
  11. 11 Grok 4.7 medium 77.1%
  12. 12 Claude Sonnet 5.5 medium 75.2%
  13. 13 Qwen 3.8 27B 73.2%
  14. 14 Claude Sonnet 5 medium 71.7%
  15. 15 Gemini 3.1 Pro medium 69%
  16. 16 Kimi K3 68.3%
  17. 17 Muse Glimmer 59.2%
  18. 18 Gemma 4 31B IT 51.9%
  19. 19 DeepSeek V4.1 Flash 49.1%
  20. 20 Nemotron 3 Nano Omni 30B-A3B 20.6%

Source: ONESTRUCTION’s leaderboard

#ModelScoreSettingDateLinkSource
1 Gemini 4 Argon 87.7% medium, pi Sep 2026 ai.onestruction.com Official leaderboard
2 GPT-6 Astra 84.3% medium, pi Sep 2026 ai.onestruction.com Official leaderboard
3 DeepSeek V4.1 Flash 1017 84.3% pi Oct 2026 ai.onestruction.com Official leaderboard
4 GPT-6.1 Sol 83.2% medium, pi Sep 2026 ai.onestruction.com Official leaderboard
5 Claude Opus 5 81.9% medium, pi Jul 2026 ai.onestruction.com Official leaderboard
6 Claude Opus 5.5 81.6% medium, pi Sep 2026 ai.onestruction.com Official leaderboard
7 Muse Spark 1.3 81.4% medium, pi Sep 2026 ai.onestruction.com Official leaderboard
8 Claude Fable 5 79.7% medium, pi Jun 2026 ai.onestruction.com Official leaderboard
9 GLM-5.3-Flash 78.9% pi Aug 2026 ai.onestruction.com Official leaderboard
10 Gemini 3.7 Flash 77.6% medium, pi Aug 2026 ai.onestruction.com Official leaderboard
11 Grok 4.7 77.1% medium, pi Sep 2026 ai.onestruction.com Official leaderboard
12 Claude Sonnet 5.5 75.2% medium, pi Sep 2026 ai.onestruction.com Official leaderboard
13 Qwen 3.8 27B 73.2% pi Aug 2026 ai.onestruction.com Official leaderboard
14 Claude Sonnet 5 71.7% medium, pi Jun 2026 ai.onestruction.com Official leaderboard
15 Gemini 3.1 Pro 69% medium, pi Feb 2026 ai.onestruction.com Official leaderboard
16 Kimi K3 68.3% pi Jul 2026 ai.onestruction.com Official leaderboard
17 Muse Glimmer 59.2% pi Aug 2026 ai.onestruction.com Official leaderboard
18 Gemma 4 31B IT 51.9% pi Apr 2026 ai.onestruction.com Official leaderboard
19 DeepSeek V4.1 Flash 49.1% pi Sep 2026 ai.onestruction.com Official leaderboard
20 Nemotron 3 Nano Omni 30B-A3B 20.6% pi Apr 2026 ai.onestruction.com Official leaderboard

The best score, 87.7%, costs $0.18 a task

Runs on the cost frontier score higher than any run that's cheaper. GPT-6.1 Sol at medium comes within 4.5 points of the best for $0.05 a task, 4 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

OpenAI Google DeepMind Cost frontier One model at each effort level
0% 20% 40% 60% 80% 100% $0.01 $0.02 $0.05 $0.10 $0.20 $0.50 $1.00 Cost per task, log scale Gemini 4 Argon, medium GPT-6.1 Sol, medium 0% 20% 40% 60% 80% 100% $0.01 $0.10 $1.00 Cost per task, log scale Gemini 4 Argon, medium

Source: ONESTRUCTION’s leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Gemini 4 Argon medium pi 87.7% $0.18
GPT-6 Astra medium pi 84.3% $0.24
DeepSeek V4.1 Flash 1017 pi 84.3%
GPT-6.1 Sol medium pi 83.2% $0.05
Claude Opus 5 medium pi 81.9% $0.49
Claude Opus 5.5 medium pi 81.6% $0.14
Muse Spark 1.3 medium pi 81.4% $0.50
Claude Fable 5 medium pi 79.7% $0.38
GLM-5.3-Flash pi 78.9%
Gemini 3.7 Flash medium pi 77.6% $0.11
Grok 4.7 medium pi 77.1% $0.26
Claude Sonnet 5.5 medium pi 75.2% $0.05
Qwen 3.8 27B pi 73.2%
Claude Sonnet 5 medium pi 71.7% $0.35
Gemini 3.1 Pro medium pi 69% $0.12
Kimi K3 pi 68.3%
Muse Glimmer pi 59.2%
Gemma 4 31B IT pi 51.9%
DeepSeek V4.1 Flash pi 49.1%
Nemotron 3 Nano Omni 30B-A3B pi 20.6%

5 models have held number one on ONESTRUCTION’s leaderboard since its release in June 2026

Lab claims count from the day the card came out, independent runs from the model’s release.

Google DeepMind Anthropic OpenAI
ONESTRUCTION leaderboard Passed 85% in Sep 2026, by Gemini 4 Argon

Source: ONESTRUCTION’s leaderboard

Gemini 4 Argon on the official leaderboard passed 85% in September 2026

That was 4 months after its release. By the same age, 24% of the benchmarks from 2026 had passed 85%. All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
On ONESTRUCTION’s leaderboard Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Gemini 4 Argon Claude Opus 5 Gemini 3.1 Pro 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Gemini 4 Argon Claude Fable 5

Source: ONESTRUCTION’s leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

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 10 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
  • Pass Independent runs Scores from the official leaderboard (20 entries).
  • Pass Models scored 20 models we track.
  • Pass Newest score The newest is from October 2026, for DeepSeek V4.1 Flash 1017.
  • Pass Frontier models 5 of the top 10 have a score, led by GPT-6 Astra (#4).
  • Note Headroom Saturated. The best is 87.7%, past 85% since September 2026.
Leaderboard import 3 pass
  • Pass Importer We read the board on ONESTRUCTION: 20 rows for models we track at the last read.
  • Pass Last read November 20th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check