Benchmark Almanac

Construction's Last Exam

Togal.ai 97.7% tops its leaderboard GPT-6.2 Astra 58.1% leads the models we track
Not in model cards
Version geometry-only

Models trace walls and areas and locate symbols on construction drawings. The score averages the three task families across a suite of 33 tasks.

Togal ranks 31 models and lists its human-assisted software separately. Eleven model scores differ between the ranking table and the progress chart.

Made by
Togal.AI
Released
October 2026
Kind
Tasks from real jobs
Level
University
Scored
Checked automatically
Answers
Agent task
Tasks
All public
Measures
Capability
Language
English
In and out
Text, Documents → Text
Interaction
Single turn
Refresh
One fixed set
Run by
Anyone
Strongest model tested
GPT-6.2 Astra #1 on the capability index
Last model added
November 2026
Scores refreshed
November 20th, 2026
32
Entries on its own leaderboard, 30 models we track

Togal.ai tops the leaderboard with 97.7% GPT-6.2 Astra leads with 58.1%

Best score per entry 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. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. The best model we track is GPT-6.2 Astra, at 58.1%. This view shows only models we track, each at its best.

  1. 1 Togal.ai Human-assisted software 97.7%
  2. 2 GPT-6.2 Astra 58.1%
  3. 3 GPT-6.2 Sol 44.2%
  4. 4 GPT-6.1 Terra 36.4%
  5. 5 GPT-6 Sol 34%
  6. 6 Gemini 4 Neon 28.3%
  7. 7 Gemini 4 Argon 25.7%
  8. 8 GPT-6 Luna 21.6%
  9. 9 GPT-6 Astra 19.2%
  10. 10 Gemini 3.8 Flash 19.1%
  11. 11 Gemini 4 Hydrogen 16.7%
  12. 12 Gemini 3.7 Flash 16.6%
  13. 13 Claude Opus 5.5 16.2%
  14. 14 Claude Opus 5 12.7%
  15. 15 Claude Fable 5.1 12.5%
  16. 16 DeepSeek Flash 4.1 12.3%
  17. 17 Gemini 3.6 Flash 12%
  18. 18 GPT-5.6 Sol 11.1%
  19. 19 GPT-5.6 Luna 10.1%
  20. 20 Gemini 3.1 Flash-Lite 6.00%
  21. 21 GLM-5.3-Flash 5.50%
  22. 22 Gemini 3.5 Flash-Lite 4.90%
  23. 23 MiniMax-M3.1-Flash 4.00%
  24. 24 Gemini 2.5 Flash Sep 2025 3.80%
  25. 25 Claude Opus 4.6 3.40%
  26. 26 Claude Haiku 4.5 3.30%
  27. 27 Claude Sonnet 4.5 2.20%
  28. 28 DeepSeek V4.1 Flash 1017 2.20%
  29. 29 Claude Sonnet 5 2.10%
  30. 30 Qwen2.5-VL-72B 1.80%
  31. 31 GLM-4.6V 1.30%
  32. 32 Qwen 3.8 Max 1.30%

Source: Official Construction's Last Exam leaderboard

#EntryScoreSettingDateSource
1 Togal.ai 97.7% Human-assisted software Official leaderboard
2 GPT-6.2 Astra 58.1% Nov 2026 Official leaderboard
3 GPT-6.2 Sol 44.2% Nov 2026 Official leaderboard
4 GPT-6.1 Terra 36.4% Oct 2026 Official leaderboard
5 GPT-6 Sol 34% Sep 2026 Official leaderboard
6 Gemini 4 Neon 28.3% Nov 2026 Official leaderboard
7 Gemini 4 Argon 25.7% Sep 2026 Official leaderboard
8 GPT-6 Luna 21.6% Sep 2026 Official leaderboard
9 GPT-6 Astra 19.2% Sep 2026 Official leaderboard
10 Gemini 3.8 Flash 19.1% Sep 2026 Official leaderboard
11 Gemini 4 Hydrogen 16.7% Nov 2026 Official leaderboard
12 Gemini 3.7 Flash 16.6% Aug 2026 Official leaderboard
13 Claude Opus 5.5 16.2% Sep 2026 Official leaderboard
14 Claude Opus 5 12.7% Jul 2026 Official leaderboard
15 Claude Fable 5.1 12.5% Sep 2026 Official leaderboard
16 DeepSeek Flash 4.1 12.3% Official leaderboard
17 Gemini 3.6 Flash 12% Jul 2026 Official leaderboard
18 GPT-5.6 Sol 11.1% Jul 2026 Official leaderboard
19 GPT-5.6 Luna 10.1% Jul 2026 Official leaderboard
20 Gemini 3.1 Flash-Lite 6.00% Mar 2026 Official leaderboard
21 GLM-5.3-Flash 5.50% Aug 2026 Official leaderboard
22 Gemini 3.5 Flash-Lite 4.90% Jul 2026 Official leaderboard
23 MiniMax-M3.1-Flash 4.00% Oct 2026 Official leaderboard
24 Gemini 2.5 Flash Sep 2025 3.80% Sep 2025 Official leaderboard
25 Claude Opus 4.6 3.40% Feb 2026 Official leaderboard
26 Claude Haiku 4.5 3.30% Oct 2025 Official leaderboard
27 Claude Sonnet 4.5 2.20% Sep 2025 Official leaderboard
28 DeepSeek V4.1 Flash 1017 2.20% Oct 2026 Official leaderboard
29 Claude Sonnet 5 2.10% Jun 2026 Official leaderboard
30 Qwen2.5-VL-72B 1.80% Jan 2025 Official leaderboard
31 GLM-4.6V 1.30% Dec 2025 Official leaderboard
32 Qwen 3.8 Max 1.30% Aug 2026 Official leaderboard
#ModelScoreSettingDateSource
1 GPT-6.2 Astra 58.1% Nov 2026 Official leaderboard
2 GPT-6.2 Sol 44.2% Nov 2026 Official leaderboard
3 GPT-6.1 Terra 36.4% Oct 2026 Official leaderboard
4 GPT-6 Sol 34% Sep 2026 Official leaderboard
5 Gemini 4 Neon 28.3% Nov 2026 Official leaderboard
6 Gemini 4 Argon 25.7% Sep 2026 Official leaderboard
7 GPT-6 Luna 21.6% Sep 2026 Official leaderboard
8 GPT-6 Astra 19.2% Sep 2026 Official leaderboard
9 Gemini 3.8 Flash 19.1% Sep 2026 Official leaderboard
10 Gemini 4 Hydrogen 16.7% Nov 2026 Official leaderboard
11 Gemini 3.7 Flash 16.6% Aug 2026 Official leaderboard
12 Claude Opus 5.5 16.2% Sep 2026 Official leaderboard
13 Claude Opus 5 12.7% Jul 2026 Official leaderboard
14 Claude Fable 5.1 12.5% Sep 2026 Official leaderboard
15 Gemini 3.6 Flash 12% Jul 2026 Official leaderboard
16 GPT-5.6 Sol 11.1% Jul 2026 Official leaderboard
17 GPT-5.6 Luna 10.1% Jul 2026 Official leaderboard
18 Gemini 3.1 Flash-Lite 6.00% Mar 2026 Official leaderboard
19 GLM-5.3-Flash 5.50% Aug 2026 Official leaderboard
20 Gemini 3.5 Flash-Lite 4.90% Jul 2026 Official leaderboard
21 MiniMax-M3.1-Flash 4.00% Oct 2026 Official leaderboard
22 Gemini 2.5 Flash Sep 2025 3.80% Sep 2025 Official leaderboard
23 Claude Opus 4.6 3.40% Feb 2026 Official leaderboard
24 Claude Haiku 4.5 3.30% Oct 2025 Official leaderboard
25 Claude Sonnet 4.5 2.20% Sep 2025 Official leaderboard
26 DeepSeek V4.1 Flash 1017 2.20% Oct 2026 Official leaderboard
27 Claude Sonnet 5 2.10% Jun 2026 Official leaderboard
28 Qwen2.5-VL-72B 1.80% Jan 2025 Official leaderboard
29 GLM-4.6V 1.30% Dec 2025 Official leaderboard
30 Qwen 3.8 Max 1.30% Aug 2026 Official leaderboard

Rankings by part

We read 3 parts from the benchmark’s own leaderboard: Linear, Areas and Counting. Each ranks models on one part of the test. GPT-6.2 Astra leads every one of them. Arrows show how far a model moved from its overall rank among the models we track.

Linear

29 models we track have a score here.

  1. 1 GPT-6.2 Astra 51.2%
  2. 2 GPT-6 Sol ↑2 30.9%
  3. 3 Gemini 4 Neon ↑2 28.5%
  4. 4 GPT-6.2 Sol ↓2 27.2%
  5. 5 Gemini 4 Argon ↑1 23.5%
  6. 6 GPT-6.1 Terra ↓3 22.1%
  7. 7 GPT-6 Astra ↑1 21.8%
  8. 8 Gemini 3.8 Flash ↑1 18.7%
  9. 9 GPT-6 Luna ↓2 15.3%
  10. 10 Claude Opus 5 ↑3 14.8%

The best score, 58.1%, costs $11 for the full run

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 Cost frontier One model at each effort level
0% 20% 40% 60% $0.10 $1.00 $10 $100 Cost of the full run, log scale GPT-6.2 Astra GPT-6.2 Sol GPT-6.1 Terra GPT-6 Luna 0% 20% 40% 60% $0.10 $1.00 $10 $100 Cost of the full run, log scale GPT-6.2 Astra GPT-6.2 Sol

Source: Official Construction's Last Exam leaderboard

ModelEffortHarnessScoreCost of the full runOutput tokens
GPT-6.2 Astra 58.1% $11
GPT-6.2 Sol 44.2% $2.80
GPT-6.1 Terra 36.4% $1.40
GPT-6 Sol 34% $3.50
Gemini 4 Neon 28.3% $4.20
Gemini 4 Argon 25.7% $2.80
GPT-6 Luna 21.6% $0.20
GPT-6 Astra 19.2% $14
Gemini 3.8 Flash 19.1% $1.20
Gemini 4 Hydrogen 16.7% $2.10
Gemini 3.7 Flash 16.6% $1.10
Claude Opus 5.5 16.2% $8.00
Claude Opus 5 12.7% $7.50
Claude Fable 5.1 12.5% $13
Gemini 3.6 Flash 12% $1.00
GPT-5.6 Sol 11.1% $5.50
GPT-5.6 Luna 10.1% $0.30
Gemini 3.1 Flash-Lite 6.00% $0.40
GLM-5.3-Flash 5.50% $0.20
Gemini 3.5 Flash-Lite 4.90% $0.60
MiniMax-M3.1-Flash 4.00% $0.20
Gemini 2.5 Flash Sep 2025 3.80% $0.50
Claude Opus 4.6 3.40% $6.50
Claude Haiku 4.5 3.30% $1.30
Claude Sonnet 4.5 2.20% $4.00
DeepSeek V4.1 Flash 1017 2.20% $0.40
Claude Sonnet 5 2.10% $3.00
Qwen2.5-VL-72B 1.80% $0.30
GLM-4.6V 1.30% $0.40
Qwen 3.8 Max 1.30% $2.20

The best score, 58.1%, costs $11 for the full run

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 Cost frontier One model at each effort level
0% 20% 40% 60% $0.10 $1.00 $10 $100 Cost of the full run, log scale GPT-6.2 Astra GPT-6.2 Sol GPT-6.1 Terra GPT-6 Luna 0% 20% 40% 60% $0.10 $1.00 $10 $100 Cost of the full run, log scale GPT-6.2 Astra GPT-6.2 Sol

Source: Official Construction's Last Exam leaderboard

ModelEffortHarnessScoreCost of the full runOutput tokens
GPT-6.2 Astra 58.1% $11
GPT-6.2 Sol 44.2% $2.80
GPT-6.1 Terra 36.4% $1.40
GPT-6 Sol 34% $3.50
Gemini 4 Neon 28.3% $4.20
Gemini 4 Argon 25.7% $2.80
GPT-6 Luna 21.6% $0.20
GPT-6 Astra 19.2% $14
Gemini 3.8 Flash 19.1% $1.20
Gemini 4 Hydrogen 16.7% $2.10
Gemini 3.7 Flash 16.6% $1.10
Claude Opus 5.5 16.2% $8.00
Claude Opus 5 12.7% $7.50
Claude Fable 5.1 12.5% $13
Gemini 3.6 Flash 12% $1.00
GPT-5.6 Sol 11.1% $5.50
GPT-5.6 Luna 10.1% $0.30
Gemini 3.1 Flash-Lite 6.00% $0.40
GLM-5.3-Flash 5.50% $0.20
Gemini 3.5 Flash-Lite 4.90% $0.60
MiniMax-M3.1-Flash 4.00% $0.20
Gemini 2.5 Flash Sep 2025 3.80% $0.50
Claude Opus 4.6 3.40% $6.50
Claude Haiku 4.5 3.30% $1.30
Claude Sonnet 4.5 2.20% $4.00
DeepSeek V4.1 Flash 1017 2.20% $0.40
Claude Sonnet 5 2.10% $3.00
Qwen2.5-VL-72B 1.80% $0.30
GLM-4.6V 1.30% $0.40
Qwen 3.8 Max 1.30% $2.20

3 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.

OpenAI
Own leaderboard 3 leaders

Source: Official Construction's Last Exam 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 Best so far
0% 20% 40% 60% 80% 100% 2025 2026 2027 Near the ceiling (85%) GPT-6.2 Astra GPT-6.1 Terra GPT-6 Astra Claude Opus 5 Gemini 3.1 Flash-Lite Gemini 2.5 Flash Qwen2.5-VL-72B 0% 20% 40% 60% 80% 100% 2025 2026 2027 Near the ceiling (85%) GPT-6.2 Astra GPT-6.1 Terra

Source: Official Construction's Last Exam 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.

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.

OpenAI Google DeepMind
Construction's Last Exam v2 Not tested since Oct 2026
Construction's Last Exam v2 Not tested since Oct 2026

Source: Official leaderboards

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

All 11 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 5 pass
  • Pass Independent runs Scores from the official leaderboard (32 entries).
  • Pass Models scored 30 models we track, plus 2 other entries on its leaderboard.
  • Pass Newest score The newest is from November 2026, for Gemini 4 Neon.
  • Pass Frontier models 6 of the top 10 have a score, led by GPT-6.2 Astra (#1).
  • Pass Headroom The best is 58.1%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on cle.togal.ai: 30 rows for models we track and 2 other entries 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