BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases9

SWE-bench Verified (any scaffold)

Best official SWE-bench Verified result for the model with any agent scaffold. Shown for reference.

As of 25 Oct 2026, Claude Opus 4.5 leads SWE-bench Verified (any scaffold) on BenchLeader with 79.2%, ahead of Gemini 3 Pro at 77.4%, across 64 model configurations with a published result.

Published by
SWE-bench
Category
Coding
Index weight
Reference only
Models
64
Data as of
25 Oct 2026
Newest result
19 Feb 2026

The publisher has posted no new results on this board since 19 Feb 2026. BenchLeader re-reads it every day, so newer models appear here as soon as they are published.

SWE-bench (Jimenez et al. 2024), swebench.com.

What the test looks like

SWE-bench Verified is 500 real GitHub issues from popular Python repositories, each screened by human annotators to make sure the task is well specified and the tests are fair. The model reads the issue, edits the repository, and its patch is accepted only if the hidden tests pass.

How it is scored

Percent of the 500 issues resolved by the model's best official submission, in whatever agent scaffold the submitter used (often the lab's own tooling), not the bash-only mini-SWE-agent harness.

What to keep in mind

The same 500 verified GitHub issues, but taking each model's best official result with any agent scaffold the submitter chose. Because scaffolds differ, this compares model-plus-harness systems rather than models alone, and the best entries often come from the labs' own tooling. Shown for reference; the bash-only board is what counts toward the index. See also the SWE-bench Verified (bash only) board.

Top 15
  1. 1Claude Opus 4.579.2%
  2. 2Gemini 3 Pro77.4%
  3. 3Claude Sonnet 476.8%
  4. 4Gemini 3 Flash (high)75.8%
  5. 5MiniMax M2.5 (high)75.8%
  6. 6Claude Opus 4.675.6%
  7. 7Claude Sonnet 4.574.8%
  8. 8GPT-574.4%
  9. 9Claude Opus 473.2%
  10. 10GPT-5.2-Codex72.8%
  11. 11GPT-5.2 (high)72.8%
  12. 12GLM-5 (high)72.8%
  13. 13Kimi K2 090571.2%
  14. 14Kimi K2.5 (high)70.8%
  15. 15Qwen3 Coder 480B69.6%
64 of 64
#
1Claude Opus 4.579.2%57.9Sonar Foundation Agent2025-12-05
2Gemini 3 Pro77.4%60.4live-SWE-agent2025-11-20
3Claude Sonnet 476.8%50.5EPAM AI/Run Developer Agent2025-08-04
4Gemini 3 Flashhigh75.8%58.3mini-SWE-agent2026-02-17
5MiniMax M2.5high75.8%–mini-SWE-agent2026-02-17
6Claude Opus 4.675.6%62.5mini-SWE-agent2026-02-17
7Claude Sonnet 4.574.8%53.7Sonar Foundation Agent2025-11-03
8GPT-574.4%56.2Prometheus-v1.2.12025-10-15
9Claude Opus 473.2%52.6Tools2025-05-22
10GPT-5.2-Codex72.8%57.3mini-SWE-agent2026-02-19
11GPT-5.2high72.8%56.0mini-SWE-agent2026-02-17
12GLM-5high72.8%–mini-SWE-agent2026-02-17
13Kimi K2 090571.2%50.5Lingxi v1.52025-10-14
14Kimi K2.5high70.8%–mini-SWE-agent2026-02-17
15Qwen3 Coder 480B69.6%48.0OpenHands2025-08-05
16GLM-4.668.2%49.9Undisclosed2025-09-30
17Claude Haiku 4.5high66.6%–mini-SWE-agent2026-02-17
18Claude 3.7 Sonnet66.4%50.0Aime-coder v12025-05-14
19GPT-5.1medium66.0%52.3mini-SWE-agent2025-11-20
20GPT-5.1-Codexmedium66.0%–mini-SWE-agent2025-11-24
21Kimi K265.4%50.4OpenHands2025-07-16
22o4-mini64.6%56.9PatchPilot-v1.12025-05-03
23o164.6%53.1W&B Programmer O1 crosscheck52025-01-17
24GLM-4.564.2%51.7Undisclosed2025-07-28
25Kimi K2 Thinkingthinking63.4%52.4mini-SWE-agent2025-12-10
26Claude 3.5 Sonnet63.4%46.3AgentScope2025-02-06
27MiniMax-M261.0%50.8mini-SWE-agent2025-11-24
284x Scaled60.8%–OpenHands2025-02-03
29Qwen3 Coder 30B A3B Instruct60.4%43.7EntroPO + R2E2025-09-01
30DeepSeek V3.2 Reasoner60.0%–mini-SWE-agent2025-12-01
31GPT-5 minimedium59.8%53.8mini-SWE-agent2025-08-07
32o358.4%59.5mini-SWE-agent2025-07-26
33Devstral Small 256.4%39.5mini-SWE-agent2025-12-09
34Devstral 253.8%42.3mini-SWE-agent2025-12-09
35Gemini 2.5 Pro53.6%53.9mini-SWE-agent2025-07-26
36Frogboss 32B 251053.6%–FrogBoss-32B-25102025-11-10
37Gemini 2.0 Flash52.2%43.2Google Jules2024-12-12
38Qwen2.5 Coder 32B Instruct47.0%42.1Skywork-SWE-32B2025-06-16
39Devstral Small 250546.8%41.7OpenHands2025-05-20
40Co-PatcheR46.0%–PatchPilot2025-05-28
41Frogmini 14B 251045.0%–FrogMini-14B-25102025-11-10
42o3-mini42.4%49.9Agentless Lite2025-02-14
43Nova Premier42.4%45.3Amazon Nova Premier 1.02025-05-27
44Llama3-SWE-RL-70B41.2%–Agentless Mini2025-02-26
45Claude 3 Haiku40.6%37.0Tools2024-10-22
46Qwen2.5-Max40.2%–SWE-agent2025-05-11
47GPT-4.139.6%48.3mini-SWE-agent2025-07-26
48GPT-4o38.8%43.9Agentless-1.52024-10-28
49Devstral Small 250738.0%41.9SWE-agent2025-07-25
50GPT-5 nanomedium34.8%45.9mini-SWE-agent2025-08-07
51Lingma SWE-GPT 72b (v0925)28.8%–Lingma Agent2024-10-02
52Gemini 2.5 Flash28.7%52.7mini-SWE-agent2025-07-26
53gpt-oss-120b26.0%45.6mini-SWE-agent2025-08-07
54GPT-4.1 mini23.9%45.0mini-SWE-agent2025-07-20
55MCTS Refine 7B23.2%–MCTS-Refine-7B2025-06-27
56GPT-4 110622.4%–SWE-agent2024-04-02
57Llama 4 Maverick21.0%42.8mini-SWE-agent2025-07-20
58Lingma SWE-GPT 7b (v0925)18.2%–Lingma Agent2024-10-02
59Claude 3 Opus15.8%39.2SWE-agent2024-04-02
60Llama 4 Scout9.1%39.7mini-SWE-agent2025-07-20

More coding benchmarks · All LLM benchmarks

Cite as: BenchLeader, “SWE-bench Verified (any scaffold) leaderboard”, data as of 25 Oct 2026, https://www.benchleader.com/benchmarks/swe_bench_verified_any.

SWE-bench Verified (any scaffold): questions

What does SWE-bench Verified (any scaffold) measure?
SWE-bench Verified is 500 real GitHub issues from popular Python repositories, each screened by human annotators to make sure the task is well specified and the tests are fair. The model reads the issue, edits the repository, and its patch is accepted only if the hidden tests pass. Scores are reported in percent of tasks solved; higher is better.
Which AI model leads SWE-bench Verified (any scaffold)?
Claude Opus 4.5 leads SWE-bench Verified (any scaffold) with 79.2% as of 25 Oct 2026, ahead of Gemini 3 Pro at 77.4%.
How many models have SWE-bench Verified (any scaffold) results?
64 model configurations have a SWE-bench Verified (any scaffold) result on BenchLeader, all taken from SWE-bench.
Who runs SWE-bench Verified (any scaffold) and how often is it updated?
SWE-bench Verified (any scaffold) is published by SWE-bench. BenchLeader re-reads the published results once a day and records the date each result was published.
Does SWE-bench Verified (any scaffold) count toward the BenchLeader Index?
No. SWE-bench Verified (any scaffold) is shown for reference but left out of the composite index.