data_distribution.md

Data Distribution

v1.0

Open-SWE-Traces v1.0 consists of 151,219 total trajectories across two LLMs and two agent scaffolds.

Table 1: Agent, Model, and PR Category-wise Distribution

Agent	Model (Type)	Category	PR by Category	Total PR	Resolved	Unresolved	Unknown	Traj. by Category	Total Traj.
OpenHands	Minimax-M2.5 (Thinking)	bug-fix enhancement feature-request other	10,719 2,083 5,180 1,040	19,022	8,725 1,415 3,619 604	9,892 2,236 6,253 1,129	5,751 1,093 2,374 512	24,368 4,744 12,246 2,245	43,603
OpenHands	Qwen3.5-122B (Non-Thinking)	bug-fix enhancement feature-request other	10,505 2,017 4,905 1,024	18,451	6,188 1,041 2,194 410	11,450 2,320 6,005 1,349	5,970 988 2,070 478	23,608 4,349 10,269 2,237	40,463
SWE-agent	Minimax-M2.5 (Thinking)	bug-fix enhancement feature-request other	11,477 2,037 4,991 1,017	19,522	10,603 1,504 3,943 699	10,633 2,091 5,416 1,066	7,005 1,040 2,291 528	28,241 4,635 11,650 2,293	46,819
SWE-agent	Qwen3.5-122B (Non-Thinking)	bug-fix enhancement feature-request other	7,926 814 1,238 533	10,511	6,147 404 533 264	5,872 599 845 434	4,245 358 400 233	16,264 1,361 1,778 931	20,334

Table 2: Language-wise Distribution

Language	OpenHands (Minimax)	OpenHands (Qwen3.5)	SWE-agent (Minimax)	SWE-agent (Qwen3.5)	Total Traj.
Python	10,522 / 4,428	8,206 / 3,939	10,899 / 4,488	4,716 / 2,284	34,343
Go	10,124 / 4,487	9,980 / 4,479	8,862 / 3,988	4,022 / 2,187	32,988
TypeScript	7,703 / 3,339	7,184 / 3,189	8,819 / 3,519	3,339 / 1,777	27,045
JavaScript	6,033 / 2,639	5,477 / 2,497	7,301 / 2,904	2,950 / 1,551	21,761
Rust	3,978 / 1,866	4,986 / 2,225	5,346 / 2,270	2,989 / 1,497	17,299
Java	2,736 / 1,203	2,536 / 1,164	3,061 / 1,278	1,180 / 629	9,513
PHP	2,305 / 971	1,890 / 864	2,292 / 971	991 / 509	7,478
C	151 / 68	143 / 64	162 / 72	85 / 48	541
C++	51 / 21	61 / 30	77 / 32	62 / 29	251
Total	43,603 / 19,022	40,463 / 18,451	46,819 / 19,522	20,334 / 10,511	151,219

---

v1.1

Open-SWE-Traces v1.1 consists of 253,182 total trajectories across two LLMs and three agent scaffolds.

Table 1: Agent, Model, and PR Category-wise Distribution

Agent	Model (Type)	Category	PR by Category	Total PR	Resolved	Unresolved	Unknown	Traj. by Category	Total Traj.
OpenHands	DeepSeek-V4-Flash-0731 (Thinking)	bug-fix enhancement feature-request other	16,061 4,241 10,068 2,290	32,660	11,725 3,116 6,540 1,729	12,318 3,584 9,298 1,985	0 0 0 0	24,043 6,700 15,838 3,714	50,295
OpenHands	Qwen3.6-27B (Non-Thinking)	bug-fix enhancement feature-request other	13,891 3,671 8,882 2,101	28,545	8,808 2,147 4,570 1,257	10,968 3,069 7,751 1,809	0 0 0 0	19,776 5,216 12,321 3,066	40,379
SWE-agent	Qwen3.6-27B (Non-Thinking)	bug-fix enhancement feature-request other	17,050 4,223 10,752 2,368	34,393	17,418 3,819 8,764 2,150	17,549 4,726 12,646 2,814	3,346 917 2,569 543	38,313 9,462 23,979 5,507	77,261
mini-SWE-agent	Qwen3.6-27B (Non-Thinking)	bug-fix enhancement feature-request other	19,443 5,028 12,286 2,783	39,540	17,655 3,826 8,737 2,013	24,794 7,036 17,876 3,916	4,361 1,166 3,199 712	46,810 12,028 29,812 6,641	95,291

Table 2: Language-wise Distribution

Language	OpenHands (DeepSeek-0731)	OpenHands (Qwen3.6)	SWE-agent (Qwen3.6)	mini-SWE-agent (Qwen3.6)	Total Traj.
Python	32,170 / 19,700	26,482 / 18,397	50,004 / 21,112	53,755 / 23,374	162,411
Go	6,013 / 4,205	4,603 / 3,336	7,124 / 3,629	13,229 / 5,059	30,969
TypeScript	4,054 / 2,881	2,999 / 2,185	6,348 / 3,047	9,264 / 3,598	22,665
JavaScript	3,045 / 2,209	2,470 / 1,771	5,006 / 2,466	6,735 / 2,807	17,256
Rust	2,542 / 1,842	1,822 / 1,377	4,289 / 2,042	6,055 / 2,301	14,708
Java	1,287 / 953	1,073 / 794	2,377 / 1,121	3,405 / 1,295	8,142
PHP	1,071 / 786	838 / 624	1,884 / 866	2,570 / 994	6,363
C	83 / 60	60 / 41	164 / 76	193 / 79	500
C++	30 / 24	32 / 20	65 / 34	85 / 33	212
Total	50,295 / 32,660	40,379 / 28,545	77,261 / 34,393	95,291 / 39,540	263,226

---

v1.2

Open-SWE-Traces v1.2 consists of 107,267 total trajectories generated by one LLM and one agent scaffold.

Table 1: Agent, Model, and PR Category-wise Distribution

Agent	Model (Type)	Category	PR by Category	Total PR	Resolved	Unresolved	Unknown	Traj. by Category	Total Traj.
mini-SWE-agent	Qwen3.8-27B (Thinking)	bug-fix enhancement feature-request other	20,836 5,166 12,544 2,859	41,405	26,770 6,068 13,537 3,550	22,363 6,024 15,254 3,029	5,044 1,292 3,574 762	54,177 13,384 32,365 7,341	107,267

Table 2: Language-wise Distribution

Language	mini-SWE-agent (Qwen3.8)	Total Traj.
Python	65,597 / 24,287	65,597
Go	13,206 / 5,338	13,206
TypeScript	9,160 / 3,743	9,160
JavaScript	7,371 / 3,047	7,371
Rust	5,804 / 2,415	5,804
Java	3,149 / 1,348	3,149
PHP	2,674 / 1,106	2,674
C	220 / 86	220
C++	86 / 35	86
Total	107,267 / 41,405	107,267

Note:

• In each Table 1, Resolved, Unresolved, and Unknown indicate issue resolution status.
• In each Table 2, counts are formatted as Trajectories / PRs.
  • PR counts are non-unique across agent/model columns and cannot be summed horizontally.
• PR counts are unique instance_id counts within each version/agent/model group and are deduplicated across configured source files.