Benchmark Almanac

Deta Studio AI Construction Takeoff

GPT-6.2 Sol 76 leads since November 2026
Not in model cards

Models use Deta’s tools to mark floor areas and count openings and electrical devices on four drawings. The rating combines accuracy, cost, and time.

Deta publishes runs for 17 distinct models. The default filter shows 15. Cost and time use estimated manual work as their reference.

Made by
Deta Studio
Released
November 2026
Kind
Tasks from real jobs
Level
University
Scored
Checked automatically
Answers
Agent task
Measures
Capability
Language
English
In and out
Text, Documents → Actions, Text
Interaction
Agentic
Environment
Deta Studio
Items from
Real work and use
Refresh
One fixed set
Run by
A host
In views
Agents
Strongest model tested
GPT-6.2 Astra #1 on the capability index
Last model added
November 2026
Scores refreshed
November 20th, 2026
17
Models scored on its own leaderboard

GPT-6.2 Sol leads with 76

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 GPT-6.2 Sol medium 76
  2. 2 GPT-6.1 Terra medium 73
  3. 3 GPT-6.2 Astra medium 73
  4. 4 GPT-6.1 Sol low 70
  5. 5 GPT-6 Astra low 63
  6. 6 Claude Fable 5.2 low 58
  7. 7 Gemini 4 Argon medium 55
  8. 8 Gemini 4 Carbon medium 55
  9. 9 Claude Opus 5.5 high 52
  10. 10 Claude Sonnet 5.5 medium 48
  11. 11 Qwen4-Max high 45
  12. 12 Kimi K3.1 Pro high 40
  13. 13 GPT-6 Luna max 34
  14. 14 Gemini 3.8 Flash high 31
  15. 15 DeepSeek V4.1 Flash 1017 high 26
  16. 16 GLM-5.3-Flash FlashX, high 15
  17. 17 Claude Haiku 5.5 low 6

Source: Official Deta Studio AI Construction Takeoff leaderboard

#ModelScoreSettingDateSource
1 GPT-6.2 Sol 76 medium Nov 2026 Official leaderboard
2 GPT-6.1 Terra 73 medium Oct 2026 Official leaderboard
3 GPT-6.2 Astra 73 medium Nov 2026 Official leaderboard
4 GPT-6.1 Sol 70 low Sep 2026 Official leaderboard
5 GPT-6 Astra 63 low Sep 2026 Official leaderboard
6 Claude Fable 5.2 58 low Nov 2026 Official leaderboard
7 Gemini 4 Argon 55 medium Sep 2026 Official leaderboard
8 Gemini 4 Carbon 55 medium Oct 2026 Official leaderboard
9 Claude Opus 5.5 52 high Sep 2026 Official leaderboard
10 Claude Sonnet 5.5 48 medium Sep 2026 Official leaderboard
11 Qwen4-Max 45 high Nov 2026 Official leaderboard
12 Kimi K3.1 Pro 40 high Oct 2026 Official leaderboard
13 GPT-6 Luna 34 max Sep 2026 Official leaderboard
14 Gemini 3.8 Flash 31 high Sep 2026 Official leaderboard
15 DeepSeek V4.1 Flash 1017 26 high Oct 2026 Official leaderboard
16 GLM-5.3-Flash 15 FlashX, high Aug 2026 Official leaderboard
17 Claude Haiku 5.5 6 low Oct 2026 Official leaderboard

Rankings by part

We read 4 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Area_Complex_1, Area_Simple_1 and Electrical_Simple_1. GPT-6 Astra leads 2 of them, and 2 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

Area_Complex_1

17 models we track have a score here.

  1. 1 Claude Fable 5.2 ↑5 99.5%
  2. 2 GPT-6 Astra ↑3 98.4%
  3. 3 GPT-6.2 Sol ↓2 98.3%
  4. 4 GPT-6.2 Astra ↓1 98.1%
  5. 5 GPT-6.1 Sol ↓1 97.9%
  6. 6 GPT-6.1 Terra ↓4 97.2%
  7. 7 Gemini 4 Carbon ↑1 97.2%
  8. 8 Gemini 4 Argon ↓1 97.1%
  9. 9 Claude Opus 5.5 93%
  10. 10 Claude Sonnet 5.5 90.9%

The best score, 76, costs $0.41 a task

Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

OpenAI Anthropic DeepSeek Cost frontier One model at each effort level
0 20 40 60 80 $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale GPT-6.2 Sol, medium GPT-6.1 Terra, low GPT-6 Luna, xhigh DeepSeek V4.1 Flash 1017, low Claude Haiku 5.5, low Claude Haiku 5.5, xhigh 0 20 40 60 80 $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale GPT-6.2 Sol, medium GPT-6.1 Terra, medium

Source: Official Deta Studio AI Construction Takeoff leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-6.2 Sol medium 76 $0.41
GPT-6.2 Sol low 74 $0.55
GPT-6.1 Terra medium 73 $0.28
GPT-6.2 Astra medium 73 $0.63
GPT-6.1 Terra high 72 $0.43
GPT-6.2 Sol high 71 $0.74
GPT-6.2 Sol xhigh 71 $0.80
GPT-6.1 Sol low 70 $0.42
GPT-6.1 Sol high 70 $0.77
GPT-6.2 Astra low 70 $0.96
GPT-6.1 Sol medium 69 $0.69
GPT-6.1 Terra xhigh 69 $0.53
GPT-6.2 Astra xhigh 68 $1.83
GPT-6.1 Sol xhigh 67 $0.89
GPT-6.1 Terra none 66 $0.29
GPT-6.2 Astra high 66 $1.50
GPT-6.1 Terra max 65 $1.03
GPT-6.2 Sol max 65 $1.32
GPT-6.1 Sol max 64 $1.30
GPT-6 Astra low 63 $3.19
GPT-6 Astra xhigh 63 $3.58
GPT-6 Astra high 60 $5.54
GPT-6 Astra medium 59 $4.56
GPT-6 Astra max 58 $5.32
Claude Fable 5.2 low 58 $15
GPT-6.2 Astra max 56 $4.18
Gemini 4 Argon medium 55 $3.09
Gemini 4 Carbon medium 55 $4.11
Gemini 4 Argon high 54 $3.93
Gemini 4 Carbon high 54 $4.94
Claude Opus 5.5 high 52 $3.52
Gemini 4 Carbon low 52 $4.67
Claude Opus 5.5 low 48 $2.38
Claude Sonnet 5.5 medium 48 $4.48
Claude Opus 5.5 xhigh 45 $2.19
Qwen4-Max high 45 $1.55
GPT-6.1 Terra low 44 $0.17
Claude Sonnet 5.5 xhigh 43 $3.61
Claude Sonnet 5.5 max 41 $1.85
Kimi K3.1 Pro high 40 $1.32
Claude Sonnet 5.5 high 38 $7.61
Claude Sonnet 5.5 low 37 $3.48
Kimi K3.1 Pro low 36 $1.28
Qwen4-Max low 36 $1.34
Kimi K3.1 Pro max 35 $2.00
GPT-6 Luna max 34 $0.21
Gemini 3.8 Flash high 31 $2.27
Claude Opus 5.5 max 28 $1.25
GPT-6 Luna xhigh 28 $0.14
Claude Opus 5.5 medium 26 $1.14
DeepSeek V4.1 Flash 1017 high 26 $0.14
Gemini 3.8 Flash low 18 $2.37
DeepSeek V4.1 Flash 1017 low 18 $0.12
GLM-5.3-Flash FlashX, high 15 $1.86
GLM-5.3-Flash FlashX, low 15 $0.99
Gemini 3.8 Flash medium 11 $1.97
GPT-6 Luna medium 10 $0.12
Qwen4-Max xhigh 9 $1.03
Claude Haiku 5.5 low 6 $0.09
GPT-6 Luna high 5 $0.13
GLM-5.3-Flash FlashX, max 4 $1.27
Claude Haiku 5.5 high 3 $0.11
Claude Haiku 5.5 medium 1 $0.06
Claude Haiku 5.5 xhigh 1 $0.04
DeepSeek V4.1 Flash 1017 max 1 $0.13
GPT-6 Luna low 0 $0.02
Claude Haiku 5.5 max 0 $0.06

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

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

OpenAI
Own leaderboard 2 leaders

Source: Official Deta Studio AI Construction Takeoff leaderboard

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
0 20 40 60 80 2027 0 20 40 60 80 2027

Source: Official Deta Studio AI Construction Takeoff 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 Economic work

All 210 in Economic work
  • JobBench In use 130 tasks from 35 occupations workers most want delegated. Agent works through messy reference files, graded by fact-anchored binary rubrics. GLM-5.3 61.4%
  • GDPval In use Real work tasks from 44 occupations with reference files. Experts blindly compare model deliverables with expert work. Metric is win-or-tie rate. Seed2.1 Pro 87.9%
  • GDPval-AA v2.1 In use Agents with shell and web access complete 220 GDPval gold tasks from 44 occupations. Blind pairwise comparisons give Elo ratings. Claude Opus 5.5 1,846
  • AA-Briefcase v1.1 In use Version 1.1 of Artificial Analysis's multi-hour office work evaluation, with Elo ratings fitted by Crowd-BT. Claude Opus 5.5 1,822
  • Agents' Last Exam v1.1 In use Agents complete 122 public long-horizon professional workflow tasks across 55 sub-industries. Deterministic graders check outcomes against hidden references. GPT-6.2 Astra 47.5%
  • CoWorkBench In use Agent completes long-horizon productivity tasks across computer science, finance, law and medicine. Qwen in-house scoring (method unpublished). Claude Fable 5 75.9%

Data audit

9 of 10 checks pass, 1 warning. Look first at the Task access check. We don’t know whether the tasks are public. Every benchmark

Catalog record 1 warning 1 pass
  • Warning Task access We don’t know whether the tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
  • Pass Independent runs Scores from the official leaderboard (17 entries).
  • Pass Models scored 17 models we track.
  • Pass Newest score The newest is from November 2026, for GPT-6.2 Astra.
  • Pass Frontier models 9 of the top 10 have a score, led by GPT-6.2 Astra (#1).
  • Note Headroom Scored as a rating, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on deta.studio: 67 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