Benchmark Almanac

Active-SWE

Claude Opus 4.8 20% leads since August 2026
Not in model cards

Coding agents inspect repository snapshots and fix bugs without issue reports. Evaluation checks patches against reference tests and measures localization precision and recall. A separate stage uses generated tests and model review to assess newly revealed bugs.

The main suite contains 400 tasks. Active-SWE-Extend contains 1,663 tasks and must not be compared as the same task set. The revealed-bug measure includes model judgment.

Released
August 2026
Kind
Fixing real repositories
Level
Expert
Scored
Checked automatically
Answers
Code
Tasks
All public
Measures
Capability
Language
English
In and out
Text, Code → Code, Actions
Interaction
Agentic
Environment
Terminal
Items from
Real work and use
Refresh
One fixed set
Run by
Anyone
In views
Agents
Strongest model tested
GPT-5.5 #21 on the capability index
Last model added
July 2026
Scores refreshed
November 20th, 2026

In the authors’ words

Coding agents powered by large language models (LLMs) are increasingly adopted in software engineering (SWE) scenarios, capable of fixing a specific bug in large-scale codebase. However, existing SWE benchmarks typically assume that high-quality issue reports with detailed information are always available, which is easily violated in practice due to the complexity of report acquisition and curation. To address this, we introduce Active-SWE, a benchmark for evaluating coding agents on proactively discovering and fixing multiple bugs without report guidance, covering 1,663 tasks across six bug categories and eight languages. Beyond shifting the focus from existing reactive bug fixing to proactive bug fixing, Active-SWE enables a more in-depth evaluation by expanding the scope from fixing a specific recorded bug to multiple-bug fixing and potential bug discovery scenarios. To construct Active-SWE, we propose a novel difficulty-aware task formulation pipeline with a dual-track evaluation framework, facilitating comprehensive evaluation of proactive bug-fixing capability. Extensive experiments reveal that most state-of-the-art coding agents struggle with proactive bug-fixing tasks, demonstrating limited performance in locating and resolving recorded bugs, handling multiple bug fixing scenarios, and discovering valid potential bugs.

Abstract of Active-SWE: Benchmarking Coding Agents for Proactive Bug Fixing without Issue Reports (arXiv:2608.04682, CC0)
17
Entries in its paper, 16 models we track

Claude Opus 4.8 leads with 20% among models released by July 2026 Claude Opus 4.8 leads with 20% among models released by July 2026

Best score per entry 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. 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. This view shows only models we track, each at its best.

The newest model with a score here came out in July 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 Claude Opus 4.8 20%
  2. 2 GPT-5.5 18.5%
  3. 3 Gemini 3.1 Pro 15.8%
  4. 4 Kimi K2.7 Code 14.8%
  5. 5 Kimi K2.6 14%
  6. 6 GLM-5.2 13%
  7. 7 Qwen3.7-Max 11.8%
  8. 8 Seed2.1 11%
  9. 9 Ring-2.6-1T 11%
  10. 10 Tencent Hy3 9.80%
  11. 11 DeepSeek-V4-Pro 9.50%
  12. 12 GPT-5.4 8.80%
  13. 13 Claude Sonnet 4.6 8.30%
  14. 14 MiniMax-M2.7 7.80%
  15. 15 Qwen3.5 397B-A17B 6.80%
  16. 16 Qwen3.5-122B-A10B 5.80%
  17. 17 Qwen3.5-35B-A3B 5.00%

Source: The Active-SWE paper

#EntryScoreSettingDateSource
1 Claude Opus 4.8 20% Claude Code May 2026 Official leaderboard
2 GPT-5.5 18.5% Claude Code Apr 2026 Official leaderboard
3 Gemini 3.1 Pro 15.8% Claude Code Feb 2026 Official leaderboard
4 Kimi K2.7 Code 14.8% Claude Code Jun 2026 Official leaderboard
5 Kimi K2.6 14% Claude Code Apr 2026 Official leaderboard
6 GLM-5.2 13% Claude Code Jun 2026 Official leaderboard
7 Qwen3.7-Max 11.8% Claude Code May 2026 Official leaderboard
8 Seed2.1 closed weights 11% Claude Code Official leaderboard
9 Ring-2.6-1T 11% Claude Code May 2026 Official leaderboard
10 Tencent Hy3 9.80% Claude Code Jul 2026 Official leaderboard
11 DeepSeek-V4-Pro 9.50% Claude Code Apr 2026 Official leaderboard
12 GPT-5.4 8.80% Claude Code Mar 2026 Official leaderboard
13 Claude Sonnet 4.6 8.30% Claude Code Feb 2026 Official leaderboard
14 MiniMax-M2.7 7.80% Claude Code Mar 2026 Official leaderboard
15 Qwen3.5 397B-A17B 6.80% Claude Code Feb 2026 Official leaderboard
16 Qwen3.5-122B-A10B 5.80% Claude Code Feb 2026 Official leaderboard
17 Qwen3.5-35B-A3B 5.00% Claude Code Feb 2026 Official leaderboard
#ModelScoreSettingDateSource
1 Claude Opus 4.8 20% Claude Code May 2026 Official leaderboard
2 GPT-5.5 18.5% Claude Code Apr 2026 Official leaderboard
3 Gemini 3.1 Pro 15.8% Claude Code Feb 2026 Official leaderboard
4 Kimi K2.7 Code 14.8% Claude Code Jun 2026 Official leaderboard
5 Kimi K2.6 14% Claude Code Apr 2026 Official leaderboard
6 GLM-5.2 13% Claude Code Jun 2026 Official leaderboard
7 Qwen3.7-Max 11.8% Claude Code May 2026 Official leaderboard
8 Ring-2.6-1T 11% Claude Code May 2026 Official leaderboard
9 Tencent Hy3 9.80% Claude Code Jul 2026 Official leaderboard
10 DeepSeek-V4-Pro 9.50% Claude Code Apr 2026 Official leaderboard
11 GPT-5.4 8.80% Claude Code Mar 2026 Official leaderboard
12 Claude Sonnet 4.6 8.30% Claude Code Feb 2026 Official leaderboard
13 MiniMax-M2.7 7.80% Claude Code Mar 2026 Official leaderboard
14 Qwen3.5 397B-A17B 6.80% Claude Code Feb 2026 Official leaderboard
15 Qwen3.5-122B-A10B 5.80% Claude Code Feb 2026 Official leaderboard
16 Qwen3.5-35B-A3B 5.00% Claude Code Feb 2026 Official leaderboard

Rankings by part

We read 2 parts from the benchmark’s paper: Simple and Hard. Each ranks models on one part of the test. Kimi K2.6 leads both. Arrows show how far a model moved from its overall rank among the models we track.

Simple

6 models we track have a score here.

  1. 1 Kimi K2.6 ↑4 13.8%
  2. 2 DeepSeek-V4-Pro ↑8 9.40%
  3. 3 Qwen3.5 397B-A17B ↑11 6.20%
  4. 4 Qwen3.5-122B-A10B ↑11 6.00%
  5. 5 Qwen3.5-35B-A3B ↑11 5.20%
  6. 6 MiniMax-M2.7 ↑7 5.10%

The best score, 20%, costs $32 a task

Runs on the cost frontier score higher than any run that's cheaper. Gemini 3.1 Pro comes within 4.2 points of the best for $7.74 a task, 4 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

Google DeepMind Moonshot AI OpenAI Ant Group Anthropic Tencent Cost frontier One model at each effort level
0% 5% 10% 15% 20% $0.10 $1.00 $10 $100 Cost per task, log scale Claude Opus 4.8 GPT-5.5 Gemini 3.1 Pro Kimi K2.6 Ring-2.6-1T Tencent Hy3 0% 5% 10% 15% 20% $0.10 $1.00 $10 $100 Cost per task, log scale Claude Opus 4.8 GPT-5.5

Source: The Active-SWE paper

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Opus 4.8 Claude Code 20% $32
GPT-5.5 Claude Code 18.5% $22
Gemini 3.1 Pro Claude Code 15.8% $7.74
Kimi K2.7 Code Claude Code 14.8% $14
Kimi K2.6 Claude Code 14% $7.53
GLM-5.2 Claude Code 13% $8.66
Qwen3.7-Max Claude Code 11.8% $13
Ring-2.6-1T Claude Code 11% $4.18
Tencent Hy3 Claude Code 9.80% $0.47
DeepSeek-V4-Pro Claude Code 9.50% $2.41
GPT-5.4 Claude Code 8.80% $5.23
Claude Sonnet 4.6 Claude Code 8.30% $12
MiniMax-M2.7 Claude Code 7.80% $2.14
Qwen3.5 397B-A17B Claude Code 6.80% $1.14
Qwen3.5-122B-A10B Claude Code 5.80% $0.74
Qwen3.5-35B-A3B Claude Code 5.00% $0.54

The best score, 20%, costs $32 a task

Runs on the cost frontier score higher than any run that's cheaper. Gemini 3.1 Pro comes within 4.2 points of the best for $7.74 a task, 4 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

Google DeepMind Moonshot AI OpenAI Ant Group Anthropic Tencent Cost frontier One model at each effort level
0% 5% 10% 15% 20% $0.10 $1.00 $10 $100 Cost per task, log scale Claude Opus 4.8 GPT-5.5 Gemini 3.1 Pro Kimi K2.6 Ring-2.6-1T Tencent Hy3 0% 5% 10% 15% 20% $0.10 $1.00 $10 $100 Cost per task, log scale Claude Opus 4.8 GPT-5.5

Source: The Active-SWE paper

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Opus 4.8 Claude Code 20% $32
GPT-5.5 Claude Code 18.5% $22
Gemini 3.1 Pro Claude Code 15.8% $7.74
Kimi K2.7 Code Claude Code 14.8% $14
Kimi K2.6 Claude Code 14% $7.53
GLM-5.2 Claude Code 13% $8.66
Qwen3.7-Max Claude Code 11.8% $13
Ring-2.6-1T Claude Code 11% $4.18
Tencent Hy3 Claude Code 9.80% $0.47
DeepSeek-V4-Pro Claude Code 9.50% $2.41
GPT-5.4 Claude Code 8.80% $5.23
Claude Sonnet 4.6 Claude Code 8.30% $12
MiniMax-M2.7 Claude Code 7.80% $2.14
Qwen3.5 397B-A17B Claude Code 6.80% $1.14
Qwen3.5-122B-A10B Claude Code 5.80% $0.74
Qwen3.5-35B-A3B Claude Code 5.00% $0.54

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

Anthropic
Benchmark paper Not tested since Aug 2026

Source: The Active-SWE paper

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
In the benchmark’s paper Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Opus 4.8 Gemini 3.1 Pro Claude Sonnet 4.6 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Opus 4.8 Claude Sonnet 4.6

Source: Official Active-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

All 523 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 (17 entries).
  • Pass Models scored 16 models we track, plus 1 other entry on its leaderboard.
  • Pass Newest score The newest is from July 2026, for Tencent Hy3.
  • Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.5 (#21).
  • Pass Headroom The best is 20%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 16 rows for models we track and 1 other entry 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