DrafterBench
Agents call drawing revision tools from written instructions. The evaluator records their operations and arguments across 1,920 tasks covering 12 types of revision.
The authors publish three-run means and variances for 19 models. Their README warns that the upgraded tasks can yield lower scores.
- Made by
- Yinsheng Li, Zhen Dong, and Yi Shao
- Released
- April 2025
- Kind
- Function calling
- Level
- University
- Scored
- Checked automatically
- Answers
- Agent task
- Tasks
- All public
- Measures
- Capability
- Subject
- Civil engineering
- Language
- English
- Interaction
- Agentic
- Environment
- APIs and tools
- Items from
- Real work and use
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Agents
- Strongest model tested
- GPT-5.4 #28 on the capability index
- Last model added
- March 2026
- Scores refreshed
- November 20th, 2026
Claude Sonnet 4.5 leads with 84.5% among models released by March 2026 Claude Sonnet 4.5 leads with 84.5% among models released by March 2026
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. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.
The newest model with a score here came out in March 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 84.5 Claude Sonnet 4.5
- 84.0 o3
- 82.6 GPT-5.4
- 82.5 GPT-5
- 80.2 Kimi K2
- 80.1
Gemini 2.5 Pro
- 78.9
DeepSeek-V3.2
- 78.3
Gemini 3.1 Flash-Lite
- 78.1 o1
- 77.9 GPT-4.1
- 76.8 o4-mini
- 75.6 Qwen3-253B-A22B-Instruct-2507
- 75.0 GPT-5.2
- 73.4 Claude 3.5 Sonnet
- 72.2 GPT-4o
- 71.7
DeepSeek-V3
- 69.9
Qwen2.5-72B
- 65.4 GPT-4o mini
- 65.0 LLaMA3-70B-Instruct
- 1 Claude Sonnet 4.5 84.5%
- 2 o3 84%
- 3 GPT-5.4 82.6%
- 4 GPT-5 82.5%
- 5 Kimi K2 Sep 2025 80.2%
- 6
Gemini 2.5 Pro Jun 2025 80.1%
- 7
DeepSeek-V3.2 78.9%
- 8
Gemini 3.1 Flash-Lite 78.3%
- 9 o1 78.1%
- 10 GPT-4.1 77.9%
- 11 o4-mini 76.8%
- 12 Qwen3-253B-A22B-Instruct-2507 75.6%
- 13 GPT-5.2 75%
- 14 Claude 3.5 Sonnet 73.4%
- 15 GPT-4o Aug 2024 72.2%
- 16
DeepSeek-V3 Mar 2025 71.7%
- 17
Qwen2.5-72B 69.9%
- 18 GPT-4o mini 65.4%
- 19 LLaMA3-70B-Instruct 65%
Source: Official DrafterBench leaderboard
- 1 Claude Sonnet 4.5 84.5%
- 2 o3 84%
- 3 GPT-5.4 82.6%
- 4 GPT-5 82.5%
- 5 Kimi K2 Sep 2025 80.2%
- 6
Gemini 2.5 Pro Jun 2025 80.1%
- 7
DeepSeek-V3.2 78.9%
- 8
Gemini 3.1 Flash-Lite 78.3%
- 9 o1 78.1%
- 10 GPT-4.1 77.9%
- 11 o4-mini 76.8%
- 12 GPT-5.2 75%
- 13 Claude 3.5 Sonnet 73.4%
- 14 GPT-4o Aug 2024 72.2%
- 15
DeepSeek-V3 Mar 2025 71.7%
- 16
Qwen2.5-72B 69.9%
- 17 GPT-4o mini 65.4%
Source: Official DrafterBench leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 84.5% | Sep 2025 | Official leaderboard | |
| 2 | o3 | 84% | Apr 2025 | Official leaderboard | |
| 3 | GPT-5.4 | 82.6% | Mar 2026 | Official leaderboard | |
| 4 | GPT-5 | 82.5% | Aug 2025 | Official leaderboard | |
| 5 | Kimi K2 Sep 2025 | 80.2% | Sep 2025 | Official leaderboard | |
| 6 | | 80.1% | Jun 2025 | Official leaderboard | |
| 7 | | 78.9% | Dec 2025 | Official leaderboard | |
| 8 | | 78.3% | Mar 2026 | Official leaderboard | |
| 9 | o1 | 78.1% | Dec 2024 | Official leaderboard | |
| 10 | GPT-4.1 | 77.9% | Apr 2025 | Official leaderboard | |
| 11 | o4-mini | 76.8% | Apr 2025 | Official leaderboard | |
| 12 | Qwen3-253B-A22B-Instruct-2507 | 75.6% | Official leaderboard | ||
| 13 | GPT-5.2 | 75% | Dec 2025 | Official leaderboard | |
| 14 | Claude 3.5 Sonnet | 73.4% | Jun 2024 | Official leaderboard | |
| 15 | GPT-4o Aug 2024 | 72.2% | Aug 2024 | Official leaderboard | |
| 16 | | 71.7% | Mar 2025 | Official leaderboard | |
| 17 | | 69.9% | Sep 2024 | Official leaderboard | |
| 18 | GPT-4o mini | 65.4% | Jul 2024 | Official leaderboard | |
| 19 | LLaMA3-70B-Instruct | 65% | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 84.5% | Sep 2025 | Official leaderboard | |
| 2 | o3 | 84% | Apr 2025 | Official leaderboard | |
| 3 | GPT-5.4 | 82.6% | Mar 2026 | Official leaderboard | |
| 4 | GPT-5 | 82.5% | Aug 2025 | Official leaderboard | |
| 5 | Kimi K2 Sep 2025 | 80.2% | Sep 2025 | Official leaderboard | |
| 6 | | 80.1% | Jun 2025 | Official leaderboard | |
| 7 | | 78.9% | Dec 2025 | Official leaderboard | |
| 8 | | 78.3% | Mar 2026 | Official leaderboard | |
| 9 | o1 | 78.1% | Dec 2024 | Official leaderboard | |
| 10 | GPT-4.1 | 77.9% | Apr 2025 | Official leaderboard | |
| 11 | o4-mini | 76.8% | Apr 2025 | Official leaderboard | |
| 12 | GPT-5.2 | 75% | Dec 2025 | Official leaderboard | |
| 13 | Claude 3.5 Sonnet | 73.4% | Jun 2024 | Official leaderboard | |
| 14 | GPT-4o Aug 2024 | 72.2% | Aug 2024 | Official leaderboard | |
| 15 | | 71.7% | Mar 2025 | Official leaderboard | |
| 16 | | 69.9% | Sep 2024 | Official leaderboard | |
| 17 | GPT-4o mini | 65.4% | Jul 2024 | Official leaderboard |
Rankings by part
We read 10 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Structured language, Unstructured language and Precise detail. o3 leads 5 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.
Structured language
17 models we track have a score here.
- 1 Claude Sonnet 4.5 87.2%
- 2 GPT-5 ↑2 85.9%
- 3 o3 ↓1 85.9%
- 4
Gemini 2.5 Pro Jun 2025 ↑2 83.2%
- 5 Kimi K2 Sep 2025 82.9%
- 6
DeepSeek-V3.2 ↑1 82.5%
- 7 GPT-5.4 ↓4 82.5%
- 8
Gemini 3.1 Flash-Lite 81.3%
- 9 o1 80.3%
- 10 o4-mini ↑1 80%
Unstructured language
17 models we track have a score here.
- 1 o3 ↑1 86.2%
- 2 Claude Sonnet 4.5 ↓1 85.8%
- 3 GPT-5 ↑1 84.3%
- 4 GPT-5.4 ↓1 83.9%
- 5 Kimi K2 Sep 2025 81.7%
- 6
Gemini 2.5 Pro Jun 2025 81.5%
- 7 o1 ↑2 80.7%
- 8 GPT-4.1 ↑2 80.4%
- 9
Gemini 3.1 Flash-Lite ↓1 80.2%
- 10
DeepSeek-V3.2 ↓3 80.1%
Precise detail
17 models we track have a score here.
- 1 o3 ↑1 91.3%
- 2 Claude Sonnet 4.5 ↓1 91.1%
- 3
Gemini 2.5 Pro Jun 2025 ↑3 91%
- 4 o1 ↑5 89.9%
- 5 GPT-5.4 ↓2 89.5%
- 6 GPT-5 ↓2 89.3%
- 7 o4-mini ↑4 88.4%
- 8
DeepSeek-V3.2 ↓1 88.2%
- 9 GPT-5.2 ↑3 87.5%
- 10 GPT-4.1 87.1%
Vague detail
17 models we track have a score here.
- 1 Claude Sonnet 4.5 81.9%
- 2 GPT-5 ↑2 80.9%
- 3 o3 ↓1 80.8%
- 4 Kimi K2 Sep 2025 ↑1 79.3%
- 5 GPT-5.4 ↓2 76%
- 6
Gemini 3.1 Flash-Lite ↑2 74.9%
- 7
DeepSeek-V3.2 74.3%
- 8
Gemini 2.5 Pro Jun 2025 ↓2 73.7%
- 9 GPT-4.1 ↑1 73.3%
- 10 o4-mini ↑1 71.5%
Complete instruction
17 models we track have a score here.
- 1 Kimi K2 Sep 2025 ↑4 89.9%
- 2 Claude Sonnet 4.5 ↓1 88.7%
- 3 o3 ↓1 87.8%
- 4
Qwen2.5-72B ↑12 87.4%
- 5 GPT-5 ↓1 86.2%
- 6
DeepSeek-V3 Mar 2025 ↑9 86.2%
- 7 GPT-5.4 ↓4 85.8%
- 8 Claude 3.5 Sonnet ↑5 84.8%
- 9
Gemini 3.1 Flash-Lite ↓1 84.5%
- 10
Gemini 2.5 Pro Jun 2025 ↓4 82%
Incomplete (error) instruction
17 models we track have a score here.
- 1 o3 ↑1 84.3%
- 2 Claude Sonnet 4.5 ↓1 84.3%
- 3 GPT-5 ↑1 84%
- 4
Gemini 2.5 Pro Jun 2025 ↑2 82.7%
- 5 o1 ↑4 82%
- 6 GPT-5.4 ↓3 81%
- 7
DeepSeek-V3.2 80.9%
- 8 o4-mini ↑3 79.7%
- 9 GPT-5.2 ↑3 78.7%
- 10 GPT-4.1 78.6%
Single object
17 models we track have a score here.
- 1 o3 ↑1 87%
- 2 Claude Sonnet 4.5 ↓1 86.4%
- 3 GPT-5 ↑1 85.4%
- 4 GPT-5.4 ↓1 83.7%
- 5 Kimi K2 Sep 2025 83.6%
- 6
Gemini 2.5 Pro Jun 2025 83.2%
- 7
Gemini 3.1 Flash-Lite ↑1 82.1%
- 8
DeepSeek-V3.2 ↓1 82.1%
- 9 o1 81.8%
- 10 o4-mini ↑1 81.1%
Multiple objects
17 models we track have a score here.
- 1 Claude Sonnet 4.5 86.6%
- 2 o3 85%
- 3 GPT-5 ↑1 84.8%
- 4 GPT-5.4 ↓1 83%
- 5
Gemini 2.5 Pro Jun 2025 ↑1 81.6%
- 6 Kimi K2 Sep 2025 ↓1 81%
- 7
DeepSeek-V3.2 80.5%
- 8
Gemini 3.1 Flash-Lite 79.4%
- 9 GPT-4.1 ↑1 79.3%
- 10 o1 ↓1 79.2%
Single operation
17 models we track have a score here.
- 1 Claude Sonnet 4.5 87.6%
- 2 GPT-5 ↑2 86.3%
- 3 o3 ↓1 86.1%
- 4 GPT-5.4 ↓1 84.5%
- 5 Kimi K2 Sep 2025 84.1%
- 6
Gemini 2.5 Pro Jun 2025 83.4%
- 7 GPT-4.1 ↑3 81.8%
- 8
Gemini 3.1 Flash-Lite 81.7%
- 9
DeepSeek-V3.2 ↓2 81.4%
- 10 o1 ↓1 81.3%
Multiple operations
17 models we track have a score here.
- 1 o3 ↑1 85.8%
- 2 Claude Sonnet 4.5 ↓1 84.6%
- 3 GPT-5 ↑1 82.1%
- 4
DeepSeek-V3.2 ↑3 81%
- 5 GPT-5.4 ↓2 80.5%
- 6
Gemini 2.5 Pro Jun 2025 79.9%
- 7 o4-mini ↑4 79.8%
- 8
Gemini 3.1 Flash-Lite 78.4%
- 9 o1 78.1%
- 10 Kimi K2 Sep 2025 ↓5 77.9%
3 models have held number one on the official leaderboard since its release in April 2025
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official DrafterBench leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official DrafterBench 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 Tool use
- AutomationBench In use Agent completes end-to-end business workflows inside a simulated company exposing dozens of REST APIs. Graded on resulting state.
Gemini 4 Neon 56.3%
- MCP Atlas In use Multi-step workflows over real MCP servers. Scores tool-use competency on realistic tasks.
Gemini 4 Argon 90.7%
- BFCL V4 In use Models call functions and tools, including multi-turn agentic tasks with web search and memory. Checked against expected calls and outcomes.
Qwen3.7-Max 75%
- τ³-bench Banking In use Agents act as banking customer-service reps over many turns with a simulated user, following policy. Pass^1 on the final database state.
Qwen 3.8 Max 55.1%
- Toolathlon-Verified In use 108 long-horizon tasks across 32 apps through about 600 MCP tools. Execution-based checkers inspect final environment state. GLM-5.3-Flash 78.4%
- DeepSearchQA In use Questions that need multi-step search and synthesis. Final answers graded against references. Kimi K3 95%
Data audit
9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 8 months ago, in March 2026, for GPT-5.4. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
- Pass Independent runs Scores from the official leaderboard (19 entries).
- Pass Models scored 17 models we track, plus 2 other entries on its leaderboard.
- Warning Newest score The newest came 8 months ago, in March 2026, for GPT-5.4.
- Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.4 (#28).
- Pass Headroom The best is 84.5%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on github.com: 17 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.