Benchmark Almanac

FrontierSWE

Claude Fable 5.2 69.9% leads since November 2026
Ling-3.1-flash 75.2% leads since October 2026
In use

34 ultra-long-horizon engineering and research tasks. 5 trials of up to 20 hours each. Score is mean task score 0-100.

This record is FrontierSWE v2 (September 2nd, 2026), the version Epoch AI tracks. It added 21 tasks to reach 34 and runs every model in the proximus harness. v1 (April 2026, 17 tasks) has its own record.

Made by
Proximal
Released
September 2026
Kind
Building whole projects
Level
Research frontier
Scored
Checked automatically
Answers
Code
Tasks
All public
Strongest model tested
Claude Fable 5.2 #1 on the capability index
Last in a launch post
November 2026
Last model added
November 2026
Scores refreshed
November 13th, 2026
25
Models scored on its own leaderboard
4
Launch posts list it, from 3 labs
8
System cards report it, from 4 labs

Claude Fable 5.2 leads with 69.9%

Best score per model from the benchmark's own leaderboard, read on November 13th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

Ling-3.1-flash leads with 75.2%, claims included

Best score per model from every source, including labs’ own numbers and claims Epoch AI collected from reports and papers. Labs pick their own settings, so a claim can beat an independent run of the same model.

25 models. Scroll sideways for the rest.

  1. 69.9 Claude Fable 5.2
  2. 65.5 GPT-6 Astra
  3. 62.3 Claude Opus 5.5
  4. 61.9 Claude Sonnet 5.5
  5. 61.2 Gemini 4 Carbon
  6. 56.3 Claude Fable 5.1
  7. 55.0 Gemini 4 Argon
  8. 53.1 Gemini 4 Hydrogen
  9. 52.0 Claude Opus 5
  10. 47.0 Claude Fable 5
  11. 43.8 Claude Haiku 5.5
  12. 32.2 GPT-5.6 Sol
  13. 30.2 GLM-5.3
  14. 29.5 Grok 4.7
  15. 25.9 Kimi K3
  16. 25.8 Muse Spark 1.3
  17. 25.3 Grok 4.6
  18. 20.3 Gemini 3.7 Flash
  19. 19.6 Gemini 3.8 Flash
  20. 18.1 GLM-5.3-Flash
  21. 17.8 Qwen3.8 Max (0902)
  22. 15.8 Qwen 3.8 Max
  23. 14.8 DeepSeek-V4-Flash-Vision-Exp
  24. 12.0 Muse Spark 1.2
  25. 4.12 Inkling
  1. 1 Claude Fable 5.2 mean@5 69.9%
  2. 2 GPT-6 Astra mean@5 65.5%
  3. 3 Claude Opus 5.5 mean@5 62.3%
  4. 4 Claude Sonnet 5.5 mean@5 61.9%
  5. 5 Gemini 4 Carbon mean@5 61.2%
  6. 6 Claude Fable 5.1 mean@5 56.3%
  7. 7 Gemini 4 Argon mean@5; Cache hit rate is much lower due to a non-production setting. 55%
  8. 8 Gemini 4 Hydrogen mean@5 53.1%
  9. 9 Claude Opus 5 mean@5 52%
  10. 10 Claude Fable 5 mean@5 47%
  11. 11 Claude Haiku 5.5 mean@5 43.8%
  12. 12 GPT-5.6 Sol mean@5 32.2%
  13. 13 GLM-5.3 mean@5 30.2%
  14. 14 Grok 4.7 mean@5 29.5%
  15. 15 Kimi K3 mean@5 25.9%
  16. 16 Muse Spark 1.3 mean@5 25.8%
  17. 17 Grok 4.6 mean@5 25.3%
  18. 18 Gemini 3.7 Flash mean@5 20.3%
  19. 19 Gemini 3.8 Flash mean@5 19.6%
  20. 20 GLM-5.3-Flash mean@5 18.1%
  21. 21 Qwen3.8 Max (0902) mean@5 17.8%
  22. 22 Qwen 3.8 Max mean@5 15.8%
  23. 23 DeepSeek-V4-Flash-Vision-Exp mean@5 14.8%
  24. 24 Muse Spark 1.2 mean@5 12%
  25. 25 Inkling mean@5 4.12%

Source: Official FrontierSWE leaderboard

27 models. Scroll sideways for the rest.

  1. 75.2 Ling-3.1-flash
  2. 69.9 Claude Fable 5.2
  3. 65.5 GPT-6 Astra
  4. 62.3 Claude Opus 5.5
  5. 61.9 Claude Sonnet 5.5
  6. 61.2 Gemini 4 Carbon
  7. 57.0 Claude Fable 5.1
  8. 55.0 Gemini 4 Argon
  9. 53.1 Gemini 4 Hydrogen
  10. 52.0 Claude Opus 5
  11. 47.0 Claude Fable 5
  12. 43.8 Claude Haiku 5.5
  13. 32.9 DeepSeek V4.1 Flash
  14. 32.2 GPT-5.6 Sol
  15. 30.2 GLM-5.3
  16. 29.5 Grok 4.7
  17. 25.9 Kimi K3
  18. 25.8 Muse Spark 1.3
  19. 25.3 Grok 4.6
  20. 20.3 Gemini 3.7 Flash
  21. 19.6 Gemini 3.8 Flash
  22. 18.1 GLM-5.3-Flash
  23. 17.8 Qwen3.8 Max (0902)
  24. 15.8 Qwen 3.8 Max
  25. 14.8 DeepSeek-V4-Flash-Vision-Exp
  26. 12.0 Muse Spark 1.2
  27. 4.12 Inkling
  1. 1 Ling-3.1-flash 75.2%
  2. 2 Claude Fable 5.2 mean@5 69.9%
  3. 3 GPT-6 Astra mean@5 65.5%
  4. 4 Claude Opus 5.5 mean@5 62.3%
  5. 5 Claude Sonnet 5.5 61.9%
  6. 6 Gemini 4 Carbon mean@5 61.2%
  7. 7 Claude Fable 5.1 57%
  8. 8 Gemini 4 Argon 55%
  9. 9 Gemini 4 Hydrogen mean@5 53.1%
  10. 10 Claude Opus 5 mean@5 52%
  11. 11 Claude Fable 5 mean@5 47%
  12. 12 Claude Haiku 5.5 43.8%
  13. 13 DeepSeek V4.1 Flash 32.9%
  14. 14 GPT-5.6 Sol mean@5 32.2%
  15. 15 GLM-5.3 mean@5 30.2%
  16. 16 Grok 4.7 mean@5 29.5%
  17. 17 Kimi K3 mean@5 25.9%
  18. 18 Muse Spark 1.3 mean@5 25.8%
  19. 19 Grok 4.6 mean@5 25.3%
  20. 20 Gemini 3.7 Flash mean@5 20.3%
  21. 21 Gemini 3.8 Flash mean@5 19.6%
  22. 22 GLM-5.3-Flash mean@5 18.1%
  23. 23 Qwen3.8 Max (0902) mean@5 17.8%
  24. 24 Qwen 3.8 Max mean@5 15.8%
  25. 25 DeepSeek-V4-Flash-Vision-Exp mean@5 14.8%
  26. 26 Muse Spark 1.2 mean@5 12%
  27. 27 Inkling mean@5 4.12%

Sources: Labs’ model cards and official FrontierSWE leaderboard

#ModelScoreSettingDateSource
1 Claude Fable 5.2 69.9% mean@5, proximus Nov 2026 Official leaderboard
2 GPT-6 Astra 65.5% mean@5, proximus Sep 2026 Official leaderboard
3 Claude Opus 5.5 62.3% mean@5, proximus Sep 2026 Official leaderboard
4 Claude Sonnet 5.5 61.9% mean@5, proximus Sep 2026 Official leaderboard
5 Gemini 4 Carbon 61.2% mean@5, proximus Oct 2026 Official leaderboard
6 Claude Fable 5.1 56.3% mean@5, proximus Sep 2026 Official leaderboard
7 Gemini 4 Argon 55% mean@5; Cache hit rate is much lower due to a non-production setting., proximus Sep 2026 Official leaderboard
8 Gemini 4 Hydrogen 53.1% mean@5, proximus Nov 2026 Official leaderboard
9 Claude Opus 5 52% mean@5, proximus Jul 2026 Official leaderboard
10 Claude Fable 5 47% mean@5, proximus Jun 2026 Official leaderboard
11 Claude Haiku 5.5 43.8% mean@5, proximus Oct 2026 Official leaderboard
12 GPT-5.6 Sol 32.2% mean@5, proximus Jul 2026 Official leaderboard
13 GLM-5.3 30.2% mean@5, proximus Aug 2026 Official leaderboard
14 Grok 4.7 29.5% mean@5, proximus Sep 2026 Official leaderboard
15 Kimi K3 25.9% mean@5, proximus Jul 2026 Official leaderboard
16 Muse Spark 1.3 25.8% mean@5, proximus Sep 2026 Official leaderboard
17 Grok 4.6 25.3% mean@5, proximus Aug 2026 Official leaderboard
18 Gemini 3.7 Flash 20.3% mean@5, proximus Aug 2026 Official leaderboard
19 Gemini 3.8 Flash 19.6% mean@5, proximus Sep 2026 Official leaderboard
20 GLM-5.3-Flash 18.1% mean@5, proximus Aug 2026 Official leaderboard
21 Qwen3.8 Max (0902) 17.8% mean@5, proximus Sep 2026 Official leaderboard
22 Qwen 3.8 Max 15.8% mean@5, proximus Aug 2026 Official leaderboard
23 DeepSeek-V4-Flash-Vision-Exp 14.8% mean@5, proximus Aug 2026 Official leaderboard
24 Muse Spark 1.2 12% mean@5, proximus Aug 2026 Official leaderboard
25 Inkling 4.12% mean@5, proximus Jul 2026 Official leaderboard
#ModelScoreSettingDateSource
1 Ling-3.1-flash 75.2% Oct 2026 Lab model card
2 Claude Fable 5.2 69.9% mean@5, proximus Nov 2026 Official leaderboard
3 GPT-6 Astra 65.5% mean@5, proximus Sep 2026 Official leaderboard
4 Claude Opus 5.5 62.3% mean@5, proximus Sep 2026 Official leaderboard
5 Claude Sonnet 5.5 61.9% Sep 2026 Lab model card
6 Gemini 4 Carbon 61.2% mean@5, proximus Oct 2026 Official leaderboard
7 Claude Fable 5.1 57% Sep 2026 Lab model card
8 Gemini 4 Argon 55% Oct 2026 Lab model card
9 Gemini 4 Hydrogen 53.1% mean@5, proximus Nov 2026 Official leaderboard
10 Claude Opus 5 52% mean@5, proximus Jul 2026 Official leaderboard
11 Claude Fable 5 47% mean@5, proximus Jun 2026 Official leaderboard
12 Claude Haiku 5.5 43.8% Oct 2026 Lab model card
13 DeepSeek V4.1 Flash 32.9% Sep 2026 Lab model card
14 GPT-5.6 Sol 32.2% mean@5, proximus Jul 2026 Official leaderboard
15 GLM-5.3 30.2% mean@5, proximus Aug 2026 Official leaderboard
16 Grok 4.7 29.5% mean@5, proximus Sep 2026 Official leaderboard
17 Kimi K3 25.9% mean@5, proximus Jul 2026 Official leaderboard
18 Muse Spark 1.3 25.8% mean@5, proximus Sep 2026 Official leaderboard
19 Grok 4.6 25.3% mean@5, proximus Aug 2026 Official leaderboard
20 Gemini 3.7 Flash 20.3% mean@5, proximus Aug 2026 Official leaderboard
21 Gemini 3.8 Flash 19.6% mean@5, proximus Sep 2026 Official leaderboard
22 GLM-5.3-Flash 18.1% mean@5, proximus Aug 2026 Official leaderboard
23 Qwen3.8 Max (0902) 17.8% mean@5, proximus Sep 2026 Official leaderboard
24 Qwen 3.8 Max 15.8% mean@5, proximus Aug 2026 Official leaderboard
25 DeepSeek-V4-Flash-Vision-Exp 14.8% mean@5, proximus Aug 2026 Official leaderboard
26 Muse Spark 1.2 12% mean@5, proximus Aug 2026 Official leaderboard
27 Inkling 4.12% mean@5, proximus Jul 2026 Official leaderboard

Labs report scores for 10 models in their cards. Switch the scores above to rank them together.

Score vs. cost

Mean score across tasks against mean cost per trial, using the leaderboard aggregation. 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% 20% 40% 60% 80% $1$10$100$1,000$10K Mean cost per trial ($), log scale, lower is better Mean@5 score Claude Fable 5 (mean@5): Mean@5 score 47%, mean cost per trial $301 GPT-5.6 Sol (mean@5): Mean@5 score 32.2%, mean cost per trial $180 Inkling (mean@5): Mean@5 score 4.12%, mean cost per trial $9.15 Kimi K3 (mean@5): Mean@5 score 25.9%, mean cost per trial $110 Claude Opus 5 (mean@5): Mean@5 score 52%, mean cost per trial $197 Qwen 3.8 Max (mean@5): Mean@5 score 15.8%, mean cost per trial $55.1 Muse Spark 1.2 (mean@5): Mean@5 score 12%, mean cost per trial $27.8 Grok 4.6 (mean@5): Mean@5 score 25.3%, mean cost per trial $243 Gemini 3.7 Flash (mean@5): Mean@5 score 20.3%, mean cost per trial $34.1 GLM-5.3 (mean@5): Mean@5 score 30.2%, mean cost per trial $97.2 Claude Fable 5.1 (mean@5): Mean@5 score 56.3%, mean cost per trial $139 Qwen3.8 Max (0902) (mean@5): Mean@5 score 17.8%, mean cost per trial $51.3 Gemini 3.8 Flash (mean@5): Mean@5 score 19.6%, mean cost per trial $38.8 Muse Spark 1.3 (mean@5): Mean@5 score 25.8%, mean cost per trial $243 GPT-6 Astra (mean@5): Mean@5 score 65.5%, mean cost per trial $1,030 Grok 4.7 (mean@5): Mean@5 score 29.5%, mean cost per trial $319 Claude Sonnet 5.5 (mean@5): Mean@5 score 61.9%, mean cost per trial $110 Gemini 4 Argon (mean@5; Cache hit rate is much lower due to a non-production setting.): Mean@5 score 55%, mean cost per trial $129 GLM-5.3-Flash (mean@5): Mean@5 score 18.1%, mean cost per trial $11 DeepSeek-V4-Flash-Vision-Exp (mean@5): Mean@5 score 14.8%, mean cost per trial $8.57 Claude Opus 5.5 (mean@5): Mean@5 score 62.3%, mean cost per trial $98.9 Claude Haiku 5.5 (mean@5): Mean@5 score 43.8%, mean cost per trial $20.6 Gemini 4 Carbon (mean@5): Mean@5 score 61.2%, mean cost per trial $61.9 Claude Fable 5.2 (mean@5): Mean@5 score 69.9%, mean cost per trial $169 Gemini 4 Hydrogen (mean@5): Mean@5 score 53.1%, mean cost per trial $61.8 Claude Fable 5.2 (mean@5)Claude Opus 5.5 (mean@5)Gemini 4 Carbon (mean@5)Gemini 4 Hydrogen (mean@5)Claude Haiku 5.5 (mean@5)GLM-5.3-Flash (mean@5)DeepSeek-V4-Flash-Vision-Exp (mean@5) 0% 20% 40% 60% 80% $1$10$100$1,000$10K Mean cost per trial ($), log scale, lower is better Mean@5 score Claude Fable 5 (mean@5): Mean@5 score 47%, mean cost per trial $301 GPT-5.6 Sol (mean@5): Mean@5 score 32.2%, mean cost per trial $180 Inkling (mean@5): Mean@5 score 4.12%, mean cost per trial $9.15 Kimi K3 (mean@5): Mean@5 score 25.9%, mean cost per trial $110 Claude Opus 5 (mean@5): Mean@5 score 52%, mean cost per trial $197 Qwen 3.8 Max (mean@5): Mean@5 score 15.8%, mean cost per trial $55.1 Muse Spark 1.2 (mean@5): Mean@5 score 12%, mean cost per trial $27.8 Grok 4.6 (mean@5): Mean@5 score 25.3%, mean cost per trial $243 Gemini 3.7 Flash (mean@5): Mean@5 score 20.3%, mean cost per trial $34.1 GLM-5.3 (mean@5): Mean@5 score 30.2%, mean cost per trial $97.2 Claude Fable 5.1 (mean@5): Mean@5 score 56.3%, mean cost per trial $139 Qwen3.8 Max (0902) (mean@5): Mean@5 score 17.8%, mean cost per trial $51.3 Gemini 3.8 Flash (mean@5): Mean@5 score 19.6%, mean cost per trial $38.8 Muse Spark 1.3 (mean@5): Mean@5 score 25.8%, mean cost per trial $243 GPT-6 Astra (mean@5): Mean@5 score 65.5%, mean cost per trial $1,030 Grok 4.7 (mean@5): Mean@5 score 29.5%, mean cost per trial $319 Claude Sonnet 5.5 (mean@5): Mean@5 score 61.9%, mean cost per trial $110 Gemini 4 Argon (mean@5; Cache hit rate is much lower due to a non-production setting.): Mean@5 score 55%, mean cost per trial $129 GLM-5.3-Flash (mean@5): Mean@5 score 18.1%, mean cost per trial $11 DeepSeek-V4-Flash-Vision-Exp (mean@5): Mean@5 score 14.8%, mean cost per trial $8.57 Claude Opus 5.5 (mean@5): Mean@5 score 62.3%, mean cost per trial $98.9 Claude Haiku 5.5 (mean@5): Mean@5 score 43.8%, mean cost per trial $20.6 Gemini 4 Carbon (mean@5): Mean@5 score 61.2%, mean cost per trial $61.9 Claude Fable 5.2 (mean@5): Mean@5 score 69.9%, mean cost per trial $169 Gemini 4 Hydrogen (mean@5): Mean@5 score 53.1%, mean cost per trial $61.8 Claude Fable 5.2 (mean@5)Claude Opus 5.5 (mean@5)Gemini 4 Carbon (mean@5)Gemini 4 Hydrogen (mean@5)Claude Haiku 5.5 (mean@5)GLM-5.3-Flash (mean@5)DeepSeek-V4-Flash-Vision-Exp (mean@5)
ModelSettingMean@5 scoreMean cost per trial
Claude Fable 5.2 mean@5, proximus 69.9% $168.8
GPT-6 Astra mean@5, proximus 65.5% $1,030
Claude Opus 5.5 mean@5, proximus 62.3% $98.87
Claude Sonnet 5.5 mean@5, proximus 61.9% $110.4
Gemini 4 Carbon mean@5, proximus 61.2% $61.93
Claude Fable 5.1 mean@5, proximus 56.3% $138.5
Gemini 4 Argon mean@5; Cache hit rate is much lower due to a non-production setting., proximus 55% $129.4
Gemini 4 Hydrogen mean@5, proximus 53.1% $61.79
Claude Opus 5 mean@5, proximus 52% $196.7
Claude Fable 5 mean@5, proximus 47% $301.3
Claude Haiku 5.5 mean@5, proximus 43.8% $20.55
GPT-5.6 Sol mean@5, proximus 32.2% $179.6
GLM-5.3 mean@5, proximus 30.2% $97.22
Grok 4.7 mean@5, proximus 29.5% $318.8
Kimi K3 mean@5, proximus 25.9% $109.7
Muse Spark 1.3 mean@5, proximus 25.8% $243.3
Grok 4.6 mean@5, proximus 25.3% $243.4
Gemini 3.7 Flash mean@5, proximus 20.3% $34.14
Gemini 3.8 Flash mean@5, proximus 19.6% $38.75
GLM-5.3-Flash mean@5, proximus 18.1% $11.03
Qwen3.8 Max (0902) mean@5, proximus 17.8% $51.27
Qwen 3.8 Max mean@5, proximus 15.8% $55.14
DeepSeek-V4-Flash-Vision-Exp mean@5, proximus 14.8% $8.57
Muse Spark 1.2 mean@5, proximus 12% $27.81
Inkling mean@5, proximus 4.1% $9.15

The best score, 69.9%, costs $169 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.

Z.ai DeepSeek Anthropic Google DeepMind Cost frontier One model at each effort level
0% 20% 40% 60% 80% $1.00 $10 $100 $1,000 $10,000 Cost per task, log scale Claude Fable 5.2, mean@5 Claude Opus 5.5, mean@5 Gemini 4 Hydrogen, mean@5 Claude Haiku 5.5, mean@5 GLM-5.3-Flash, mean@5 DeepSeek-V4-Flash-Vision-Exp, mean@5 0% 20% 40% 60% 80% $1.00 $10 $100 $1,000 $10,000 Cost per task, log scale Claude Fable 5.2, mean@5 Claude Opus 5.5, mean@5

Source: Official FrontierSWE leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Claude Fable 5.2 mean@5 proximus 69.9% $169
GPT-6 Astra mean@5 proximus 65.5% $1,030
Claude Opus 5.5 mean@5 proximus 62.3% $99
Claude Sonnet 5.5 mean@5 proximus 61.9% $110
Gemini 4 Carbon mean@5 proximus 61.2% $62
Claude Fable 5.1 mean@5 proximus 56.3% $139
Gemini 4 Argon mean@5; Cache hit rate is much lower due to a non-production setting. proximus 55% $129
Gemini 4 Hydrogen mean@5 proximus 53.1% $62
Claude Opus 5 mean@5 proximus 52% $197
Claude Fable 5 mean@5 proximus 47% $301
Claude Haiku 5.5 mean@5 proximus 43.8% $21
GPT-5.6 Sol mean@5 proximus 32.2% $180
GLM-5.3 mean@5 proximus 30.2% $97
Grok 4.7 mean@5 proximus 29.5% $319
Kimi K3 mean@5 proximus 25.9% $110
Muse Spark 1.3 mean@5 proximus 25.8% $243
Grok 4.6 mean@5 proximus 25.3% $243
Gemini 3.7 Flash mean@5 proximus 20.3% $34
Gemini 3.8 Flash mean@5 proximus 19.6% $39
GLM-5.3-Flash mean@5 proximus 18.1% $11
Qwen3.8 Max (0902) mean@5 proximus 17.8% $51
Qwen 3.8 Max mean@5 proximus 15.8% $55
DeepSeek-V4-Flash-Vision-Exp mean@5 proximus 14.8% $8.57
Muse Spark 1.2 mean@5 proximus 12% $28
Inkling mean@5 proximus 4.12% $9.15

3 models have held number one on the official leaderboard since its release in September 2026 3 models have held number one with claims included since its release in September 2026

Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard and what labs claim in their cards.

Anthropic OpenAI Ant Group
Own leaderboard 3 leaders
Claude Fable 5.2 69.9%
Lab claims 3 leaders
Ling-3.1-flash 75.2%

Sources: Official FrontierSWE leaderboard and labs’ model cards

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 Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Ling-3.1-flash GPT-6 Astra Claude Fable 5.1 Claude Opus 5 Claude Fable 5 0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Ling-3.1-flash GPT-6 Astra

Sources: Official FrontierSWE leaderboard and labs’ model cards

3 labs put it in a launch post in the last six months

Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.

Google DeepMind
Ant Group
Anthropic
DateLaunch postScoreAs the post writes it
Nov 9, 2026 Claude Fable 5.2 69.9 Agentic coding FrontierSWE v2, max effort (launch table and Table 8.1.A)
Oct 28, 2026 Gemini 4 Carbon 61.2 FrontierSWE 2
Oct 13, 2026 Ling-3.1-flash 75.2 FrontierSWE
Oct 13, 2026 Gemini 4 Argon 55 FrontierSWE v2

8 system cards report it, from 4 labs

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

DateDocumentScoreAs the document writes itSection
Nov 9, 2026 Claude Fable 5.2 system card 69.9 FrontierSWE v2, max effort, Proximal harness, mean over five trials per task, 20-hour budget (Section 8.7) Capabilities
Oct 13, 2026 Gemini 4 Argon model card 55 FrontierSWE v2 (results table, p.5) Capabilities
Oct 7, 2026 Claude Haiku 5.5 system card 43.8 FrontierSWE v2, max effort, Proximal harness, five trials per task (8.6) Capabilities
Sep 28, 2026 Claude Sonnet 5.5 system card 61.9 FrontierSWE v2, mean of five trials, Proximal harness at max effort (8.7) Capabilities
Sep 22, 2026 Claude Opus 5.5 system card 62.3 FrontierSWE v2, mean of five trials per task, Proximal harness at max effort (8.7) Capabilities
Sep 21, 2026 Grok 4.7 model card 29 FrontierSWE V2, mean@5 partial-credit score, xhigh, run by Proximal Labs (2.4) Capabilities
Sep 10, 2026 DeepSeek V4.1 Flash technical report 32.9 FrontierSWE v2, no-GPU public subset, Mean@5, multi-agent Agent Team mode at 20h deadline (Figure 10 text). Single-agent 28.2 Capabilities
Sep 1, 2026 Claude Fable 5.1 system card 57 FrontierSWE v2, mean score 0.57 on a 0 to 1 scale, max effort, Proximal harness, 5 trials per task (8.5) Capabilities

The lab's number comes out ahead in 1 of 2 pairs that differ

Each score a lab reported for its own model, beside the same model's best score on the leaderboard. Gaps mostly come from different harnesses, effort levels and task subsets. 6 of them match, which usually means one side copied the other. Every benchmark.

Model Lab says Board says Gap
Claude Fable 5.1 57% FrontierSWE v2, mean score 0.57 on a 0 to 1 scale, max effort, Proximal harness, 5 trials per task (8.5) 56.3% Leaderboard, mean@5, proximus +0.7 pts
Grok 4.7 29% FrontierSWE V2, mean@5 partial-credit score, xhigh, run by Proximal Labs (2.4) 29.5% Leaderboard, mean@5, proximus −0.5 pts
Gemini 4 Argon 55% FrontierSWE v2 55% Leaderboard, mean@5; Cache hit rate is much lower due to a non-production setting., proximus Same
Claude Fable 5.2 69.9% Agentic coding FrontierSWE v2, max effort (launch table and Table 8.1.A) 69.9% Leaderboard, mean@5, proximus Same
Claude Opus 5.5 62.3% FrontierSWE v2, mean of five trials per task, Proximal harness at max effort (8.7) 62.3% Leaderboard, mean@5, proximus Same
Claude Haiku 5.5 43.8% FrontierSWE v2, max effort, Proximal harness, five trials per task (8.6) 43.8% Leaderboard, mean@5, proximus Same
Claude Sonnet 5.5 61.9% FrontierSWE v2, mean of five trials, Proximal harness at max effort (8.7) 61.9% Leaderboard, mean@5, proximus Same
Gemini 4 Carbon 61.2% FrontierSWE 2 61.2% Leaderboard, mean@5, proximus Same

Versions

We switch the top row to each new version when it comes out. Open it for one row per version. An old version stays in play while people still test it, so whoever first takes it past 85% gets credit for closing it out.

Anthropic OpenAI
FrontierSWE 8 leaders
Claude Fable 5.2 69.9%
FrontierSWE v1 Passed 85% in Jun 2026, by Claude Fable 5
Claude Fable 5 88.2%
FrontierSWE 3 leaders
Claude Fable 5.2 69.9%

Source: Official leaderboards

Replaces FrontierSWE v1

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