Benchmark Almanac

SWE-bench CLI

Claude Opus 5.5 16.6% leads since October 2026
Not in model cards

Agents make verified multi-file changes in unfamiliar open-source repositories across 200 tasks, 11 languages and 10 technical domains. Harbor runs fail-to-pass and pass-to-pass test suites on the result.

Not part of the Princeton SWE-bench family. The board reports Pass@1.

Made by
Snorkel AI
Released
October 2026
Kind
Fixing real repositories
Level
Research frontier
Scored
Checked automatically
Answers
Agent task
Tasks
Kept private
Strongest model tested
GPT-6 Astra #3 on the capability index
Last model added
October 2026
Scores refreshed
November 12th, 2026
16
Models scored on its own leaderboard

Claude Opus 5.5 leads with 16.6%

Best score per model from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

16 models. Scroll sideways for the rest.

  1. 16.6 Claude Opus 5.5
  2. 16.3 Gemini 4 Argon
  3. 14.5 Claude Fable 5.1
  4. 14.0 Claude Opus 5
  5. 12.4 GPT-6 Astra
  6. 11.4 Muse Wonder
  7. 9.80 GPT-6.1 Sol
  8. 6.30 Gemini 3.8 Flash
  9. 6.10 Grok 4.7
  10. 5.70 Grok 4.6
  11. 4.90 Qwen 3.8 Max
  12. 4.10 Kimi K3
  13. 3.50 GLM-5.3
  14. 2.30 DeepSeek-V4-Pro
  15. 2.20 Muse Spark 1.3
  16. 0.50 Nemotron 3 Ultra
  1. 1 Claude Opus 5.5 16.6%
  2. 2 Gemini 4 Argon 16.3%
  3. 3 Claude Fable 5.1 14.5%
  4. 4 Claude Opus 5 14%
  5. 5 GPT-6 Astra 12.4%
  6. 6 Muse Wonder 11.4%
  7. 7 GPT-6.1 Sol 9.80%
  8. 8 Gemini 3.8 Flash 6.30%
  9. 9 Grok 4.7 6.10%
  10. 10 Grok 4.6 5.70%
  11. 11 Qwen 3.8 Max 4.90%
  12. 12 Kimi K3 4.10%
  13. 13 GLM-5.3 3.50%
  14. 14 DeepSeek-V4-Pro 2.30%
  15. 15 Muse Spark 1.3 2.20%
  16. 16 Nemotron 3 Ultra 0.50%

Source: Official SWE-bench CLI leaderboard

#ModelScoreSettingDateSource
1 Claude Opus 5.5 16.6% Sep 2026 Official leaderboard
2 Gemini 4 Argon 16.3% Sep 2026 Official leaderboard
3 Claude Fable 5.1 14.5% Sep 2026 Official leaderboard
4 Claude Opus 5 14% Jul 2026 Official leaderboard
5 GPT-6 Astra 12.4% Sep 2026 Official leaderboard
6 Muse Wonder 11.4% Oct 2026 Official leaderboard
7 GPT-6.1 Sol 9.80% Sep 2026 Official leaderboard
8 Gemini 3.8 Flash 6.30% Sep 2026 Official leaderboard
9 Grok 4.7 6.10% Sep 2026 Official leaderboard
10 Grok 4.6 5.70% Aug 2026 Official leaderboard
11 Qwen 3.8 Max 4.90% Aug 2026 Official leaderboard
12 Kimi K3 4.10% Jul 2026 Official leaderboard
13 GLM-5.3 3.50% Aug 2026 Official leaderboard
14 DeepSeek-V4-Pro 2.30% Apr 2026 Official leaderboard
15 Muse Spark 1.3 2.20% Sep 2026 Official leaderboard
16 Nemotron 3 Ultra 0.50% Jun 2026 Official leaderboard

Pass@1 vs. cost per trial

The line joins the runs that no other run beats on both measures. We redrew this chart from the numbers on the official leaderboard.

0% 5% 10% 15% 20% $0$5$10$15$20$25 Cost per trial ($), lower is better Pass@1 DeepSeek-V4-Pro: Pass@1 2.3%, cost per trial $2.2 Nemotron 3 Ultra: Pass@1 0.5%, cost per trial $4.62 Kimi K3: Pass@1 4.1%, cost per trial $6.09 Claude Opus 5: Pass@1 14%, cost per trial $23.7 Qwen 3.8 Max: Pass@1 4.9%, cost per trial $3.15 Grok 4.6: Pass@1 5.7%, cost per trial $3.3 GLM-5.3: Pass@1 3.5%, cost per trial $8.19 Claude Fable 5.1: Pass@1 14.5%, cost per trial $13 Gemini 3.8 Flash: Pass@1 6.3%, cost per trial $4.61 Muse Spark 1.3: Pass@1 2.2%, cost per trial $5.09 Grok 4.7: Pass@1 6.1%, cost per trial $5.2 Muse Wonder: Pass@1 11.4%, cost per trial $17.9 GPT-6 Astra: Pass@1 12.4%, cost per trial $2.16 Claude Opus 5.5: Pass@1 16.6%, cost per trial $3.17 GPT-6.1 Sol: Pass@1 9.8%, cost per trial $0.29 Gemini 4 Argon: Pass@1 16.3%, cost per trial $2.49 Claude Opus 5.5Gemini 4 ArgonGPT-6 AstraGPT-6.1 Sol 0% 5% 10% 15% 20% $0$5$10$15$20$25 Cost per trial ($), lower is better Pass@1 DeepSeek-V4-Pro: Pass@1 2.3%, cost per trial $2.2 Nemotron 3 Ultra: Pass@1 0.5%, cost per trial $4.62 Kimi K3: Pass@1 4.1%, cost per trial $6.09 Claude Opus 5: Pass@1 14%, cost per trial $23.7 Qwen 3.8 Max: Pass@1 4.9%, cost per trial $3.15 Grok 4.6: Pass@1 5.7%, cost per trial $3.3 GLM-5.3: Pass@1 3.5%, cost per trial $8.19 Claude Fable 5.1: Pass@1 14.5%, cost per trial $13 Gemini 3.8 Flash: Pass@1 6.3%, cost per trial $4.61 Muse Spark 1.3: Pass@1 2.2%, cost per trial $5.09 Grok 4.7: Pass@1 6.1%, cost per trial $5.2 Muse Wonder: Pass@1 11.4%, cost per trial $17.9 GPT-6 Astra: Pass@1 12.4%, cost per trial $2.16 Claude Opus 5.5: Pass@1 16.6%, cost per trial $3.17 GPT-6.1 Sol: Pass@1 9.8%, cost per trial $0.29 Gemini 4 Argon: Pass@1 16.3%, cost per trial $2.49 Claude Opus 5.5Gemini 4 ArgonGPT-6 AstraGPT-6.1 Sol
ModelSettingPass@1Cost per trial
Claude Opus 5.5 16.6% $3.17
Gemini 4 Argon 16.3% $2.49
Claude Fable 5.1 14.5% $12.99
Claude Opus 5 14% $23.73
GPT-6 Astra 12.4% $2.16
Muse Wonder 11.4% $17.86
GPT-6.1 Sol 9.8% $0.29
Gemini 3.8 Flash 6.3% $4.61
Grok 4.7 6.1% $5.2
Grok 4.6 5.7% $3.3
Qwen 3.8 Max 4.9% $3.15
Kimi K3 4.1% $6.09
GLM-5.3 3.5% $8.19
DeepSeek-V4-Pro 2.3% $2.2
Muse Spark 1.3 2.2% $5.09
Nemotron 3 Ultra 0.5% $4.62

The best score, 16.6%, costs $3.17 a task

Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

OpenAI Anthropic Google DeepMind 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 5.5 GPT-6 Astra GPT-6.1 Sol 0% 5% 10% 15% 20% $0.10 $1.00 $10 $100 Cost per task, log scale Claude Opus 5.5 GPT-6 Astra

Source: Official SWE-bench CLI leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Opus 5.5 16.6% $3.17
Gemini 4 Argon 16.3% $2.49
Claude Fable 5.1 14.5% $13
Claude Opus 5 14% $24
GPT-6 Astra 12.4% $2.16
Muse Wonder 11.4% $18
GPT-6.1 Sol 9.80% $0.29
Gemini 3.8 Flash 6.30% $4.61
Grok 4.7 6.10% $5.20
Grok 4.6 5.70% $3.30
Qwen 3.8 Max 4.90% $3.15
Kimi K3 4.10% $6.09
GLM-5.3 3.50% $8.19
DeepSeek-V4-Pro 2.30% $2.20
Muse Spark 1.3 2.20% $5.09
Nemotron 3 Ultra 0.50% $4.62

Number one over time

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

Anthropic
Own leaderboard 1 leader
Claude Opus 5.5 16.6%

Source: Official SWE-bench CLI leaderboard

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Opus 5.5 Claude Opus 5 Kimi K3 DeepSeek-V4-Pro 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Opus 5.5 Kimi K3

Source: Official SWE-bench CLI leaderboard

No lab model card reports it yet

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark.

More in Coding

All 282 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