Longtail-SWE
Agents resolve 202 GitHub issues across 13 programming languages with limited benchmark coverage. Hidden tests check the fix and verify that previously passing behavior still works.
The November 11, 2026 paper compares nine model configurations. Claude Opus 5.5 led at 81.52% pass@1, followed by GPT-6.1 Sol at 78.38%.
- Made by
- Vladimir Dobrovolskii and colleagues
- Released
- November 2026
- Kind
- Fixing real repositories
- Level
- Expert
- Scored
- Checked automatically
- Answers
- Agent task
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Agentic
- Environment
- Terminal
- Items from
- Real work and use
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Agents
- Strongest model tested
- Claude Opus 5.5 #5 on the capability index
- Last model added
- September 2026
- Scores refreshed
- November 20th, 2026
Claude Opus 5.5 leads with 81.5% among models released by September 2026
Best score per model from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best.
The newest model with a score here came out in September 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-6.1 Terra
- and 13 more
- 1 Claude Opus 5.5 high, 1M context 81.5%
- 2 GPT-6.1 Sol high, 1M context 78.4%
- 3 GPT-5.6 Sol high, 1M context 75.7%
- 4 GPT-5.6 Luna xhigh, 1M context 71.5%
- 5 Kimi K3 max, 262K context 63.7%
- 6
DeepSeek-V4-Flash high, 1M context 56.1%
- 7 GLM-5.3-Flash max, 262K context 48.2%
- 8
Qwen 3.8 27B xhigh, 262K context 42.7%
- 9
Qwen3.6 27B thinking, 262K context 16.7%
Source: The Longtail-SWE paper
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 | 81.5% | high, 1M context, SWE-agent-like scaffold | Sep 2026 | Official leaderboard |
| 2 | GPT-6.1 Sol | 78.4% | high, 1M context, SWE-agent-like scaffold | Sep 2026 | Official leaderboard |
| 3 | GPT-5.6 Sol | 75.7% | high, 1M context, SWE-agent-like scaffold | Jul 2026 | Official leaderboard |
| 4 | GPT-5.6 Luna | 71.5% | xhigh, 1M context, SWE-agent-like scaffold | Jul 2026 | Official leaderboard |
| 5 | Kimi K3 | 63.7% | max, 262K context, SWE-agent-like scaffold | Jul 2026 | Official leaderboard |
| 6 | | 56.1% | high, 1M context, SWE-agent-like scaffold | Apr 2026 | Official leaderboard |
| 7 | GLM-5.3-Flash | 48.2% | max, 262K context, SWE-agent-like scaffold | Aug 2026 | Official leaderboard |
| 8 | | 42.7% | xhigh, 262K context, SWE-agent-like scaffold | Aug 2026 | Official leaderboard |
| 9 | | 16.7% | thinking, 262K context, SWE-agent-like scaffold | Apr 2026 | Official leaderboard |
Rankings by part
We read 13 parts from the benchmark’s paper, each ranking models on one part of the test, such as Clojure, Dart and Elixir. Claude Opus 5.5 leads 5 of them, and 4 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
Clojure
9 models we track have a score here.
- 1 Claude Opus 5.5 82.1%
- 2 GPT-6.1 Sol 74.4%
- 3 Kimi K3 ↑2 64.1%
- 4 GPT-5.6 Sol ↓1 51.3%
- 5 GPT-5.6 Luna ↓1 46.2%
- 6
Qwen 3.8 27B ↑2 38.5%
- 7
DeepSeek-V4-Flash ↓1 35.9%
- 8 GLM-5.3-Flash ↓1 30.8%
- 9
Qwen3.6 27B 10.3%
Dart
9 models we track have a score here.
- 1 GPT-5.6 Luna ↑3 81%
- 2 GPT-5.6 Sol ↑1 81%
- 3 Claude Opus 5.5 ↓2 71.4%
- 4 Kimi K3 ↑1 66.7%
- 5
DeepSeek-V4-Flash ↑1 61.9%
- 6 GPT-6.1 Sol ↓4 61.9%
- 7 GLM-5.3-Flash 40.5%
- 8
Qwen 3.8 27B 38.1%
- 9
Qwen3.6 27B 11.9%
Elixir
9 models we track have a score here.
- 1 GPT-6.1 Sol ↑1 81.2%
- 2 Claude Opus 5.5 ↓1 75.4%
- 3 GPT-5.6 Sol 72.5%
- 4 GPT-5.6 Luna 65.2%
- 5 Kimi K3 53.6%
- 6
DeepSeek-V4-Flash 34.8%
- 7 GLM-5.3-Flash 31.9%
- 8
Qwen 3.8 27B 30.4%
- 9
Qwen3.6 27B 8.70%
Erlang
9 models we track have a score here.
- 1 Claude Opus 5.5 100%
- 2 GPT-6.1 Sol 100%
- 3 GPT-5.6 Sol 86.1%
- 4 GPT-5.6 Luna 83.3%
- 5 Kimi K3 83.3%
- 6
DeepSeek-V4-Flash 75%
- 7
Qwen 3.8 27B ↑1 58.3%
- 8 GLM-5.3-Flash ↓1 50%
- 9
Qwen3.6 27B 25%
F#
9 models we track have a score here.
- 1 GPT-5.6 Sol ↑2 81.3%
- 2 GPT-5.6 Luna ↑2 68.8%
- 3 Claude Opus 5.5 ↓2 66.7%
- 4 GPT-6.1 Sol ↓2 62.5%
- 5 Kimi K3 47.9%
- 6 GLM-5.3-Flash ↑1 43.8%
- 7
DeepSeek-V4-Flash ↓1 41.7%
- 8
Qwen 3.8 27B 29.2%
- 9
Qwen3.6 27B 10.4%
Haskell
9 models we track have a score here.
- 1 GPT-6.1 Sol ↑1 95.2%
- 2 Claude Opus 5.5 ↓1 92.9%
- 3 GPT-5.6 Luna ↑1 83.3%
- 4 GPT-5.6 Sol ↓1 83.3%
- 5
DeepSeek-V4-Flash ↑1 78.6%
- 6 Kimi K3 ↓1 76.2%
- 7 GLM-5.3-Flash 66.7%
- 8
Qwen 3.8 27B 52.4%
- 9
Qwen3.6 27B 26.2%
Julia
9 models we track have a score here.
- 1 GPT-6.1 Sol ↑1 93.8%
- 2 Claude Opus 5.5 ↓1 86.4%
- 3 GPT-5.6 Sol 85.2%
- 4 GPT-5.6 Luna 77.8%
- 5 Kimi K3 76.5%
- 6 GLM-5.3-Flash ↑1 59.3%
- 7
DeepSeek-V4-Flash ↓1 58%
- 8
Qwen 3.8 27B 38.3%
- 9
Qwen3.6 27B 16%
Kotlin
9 models we track have a score here.
- 1 Claude Opus 5.5 77.8%
- 2 GPT-5.6 Sol ↑1 72.2%
- 3
DeepSeek-V4-Flash ↑3 66.7%
- 4 GPT-6.1 Sol ↓2 63.9%
- 5 GPT-5.6 Luna ↓1 61.1%
- 6 GLM-5.3-Flash ↑1 58.3%
- 7 Kimi K3 ↓2 50%
- 8
Qwen 3.8 27B 47.2%
- 9
Qwen3.6 27B 22.2%
Lua
9 models we track have a score here.
- 1 GPT-5.6 Sol ↑2 89.7%
- 2 Claude Opus 5.5 ↓1 89.7%
- 3 GPT-6.1 Sol ↓1 89.7%
- 4 GPT-5.6 Luna 74.4%
- 5
DeepSeek-V4-Flash ↑1 66.7%
- 6 Kimi K3 ↓1 66.7%
- 7 GLM-5.3-Flash 59%
- 8
Qwen 3.8 27B 51.3%
- 9
Qwen3.6 27B 17.9%
OCaml
9 models we track have a score here.
- 1 Claude Opus 5.5 84.6%
- 2 GPT-5.6 Luna ↑2 76.9%
- 3 GPT-6.1 Sol ↓1 76.9%
- 4 GPT-5.6 Sol ↓1 64.1%
- 5 Kimi K3 56.4%
- 6
DeepSeek-V4-Flash 53.8%
- 7 GLM-5.3-Flash 43.6%
- 8
Qwen 3.8 27B 28.2%
- 9
Qwen3.6 27B 5.10%
R
9 models we track have a score here.
- 1 Kimi K3 ↑4 81%
- 2 Claude Opus 5.5 ↓1 78.6%
- 3 GPT-6.1 Sol ↓1 78.6%
- 4 GPT-5.6 Sol ↓1 76.2%
- 5
DeepSeek-V4-Flash ↑1 69%
- 6 GPT-5.6 Luna ↓2 69%
- 7
Qwen 3.8 27B ↑1 69%
- 8 GLM-5.3-Flash ↓1 57.1%
- 9
Qwen3.6 27B 42.9%
Scala
9 models we track have a score here.
- 1 Claude Opus 5.5 87%
- 2 GPT-5.6 Luna ↑2 70.4%
- 3 GPT-5.6 Sol 66.7%
- 4 GPT-6.1 Sol ↓2 66.7%
- 5 Kimi K3 53.7%
- 6
DeepSeek-V4-Flash 50%
- 7
Qwen 3.8 27B ↑1 44.4%
- 8 GLM-5.3-Flash ↓1 37%
- 9
Qwen3.6 27B 14.8%
Swift
9 models we track have a score here.
- 1 GPT-5.6 Luna ↑3 69.2%
- 2 GPT-5.6 Sol ↑1 69.2%
- 3 Claude Opus 5.5 ↓2 69.2%
- 4 GPT-6.1 Sol ↓2 64.1%
- 5
DeepSeek-V4-Flash ↑1 56.4%
- 6 GLM-5.3-Flash ↑1 53.8%
- 7 Kimi K3 ↓2 51.3%
- 8
Qwen 3.8 27B 46.2%
- 9
Qwen3.6 27B 12.8%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The Longtail-SWE paper
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official Longtail-SWE 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 Coding
- SWE-bench Verified In use 500 human-validated GitHub issues from 12 Python repos. Agent writes a patch that must pass hidden tests. Claude Opus 4.7 83.5%
- LiveCodeBench v6 In use Competitive programming problems collected after model cutoffs from LeetCode, AtCoder, Codeforces. Scored pass@1 on hidden tests.
DeepSeek-V4-Pro 93.5%
- SWE-bench Multilingual In use 300 GitHub-issue tasks across nine programming languages in SWE-bench format. Patch graded by repository tests. Claude Fable 5.2 96.6%
- DeepSWE In use 113 original long-horizon engineering tasks across 91 repos and five languages, written from scratch. Hand-written verifiers accept alternative solutions. GPT-6 Astra 74.1%
- SciCode In use Scientist-curated research coding problems split into subproblems across natural sciences. Code checked against test cases.
Seed2.1 Pro 59.8%
- Codeforces rating In use Model solves recent Codeforces contest problems. An Elo-style rating is computed from simulated contest performance.
DeepSeek V4.1 Flash 3,471
Data audit
All 10 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
- Pass Independent runs Scores from the benchmark’s paper (9 entries).
- Pass Models scored 9 models we track.
- Pass Newest score The newest is from September 2026, for GPT-6.1 Sol.
- Note Frontier models Only 2 of the top 10 have a score. The strongest tested is Claude Opus 5.5 (#5).
- Pass Headroom The best is 81.5%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 9 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.