Benchmark Almanac

Aider Polyglot

GPT-5 88% leads since Aug 2025 Fading

Models edit code to solve 225 hard Exercism exercises in C++, Go, Java, JavaScript, Python, and Rust. Scored by the share passing unit tests.

Exercises come from public Exercism tracks, so contamination is possible.

Made by Aider Released Dec 2024 Function-level problems All tasks public Last in a launch post Feb 2026 , dropped after it saturated Last model added Feb 2026 Scores refreshed 2 Nov 2026
Level
University Undergraduates and entry-level professionals
Scored
Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
Answers
Code A program or patch run against tests.
Aider-AI/polyglot-benchmark 227 stars Leaderboard aider.chat
54
Models scored in Epoch AI data
26
Launch posts list it, from 7 labs
17
System cards report it, from 9 labs

GPT-5 leads with 88% among models released by Dec 2025

Best score per model in Epoch AI’s data. Where a model ran with several settings or harnesses, we keep its best.

The newest model with a score here came out in Dec 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

54 models. Scroll sideways for the rest.

  1. 88.0 GPT-5
  2. 84.9 o3-pro
  3. 83.1 Gemini 2.5 Pro
  4. 81.3 o3
  5. 79.6 Grok 4
  6. 76.9 Gemini 2.5 Pro
  7. 74.2 DeepSeek-V3.2-Exp
  8. 74.2 DeepSeek-V3.2
  9. 72.9 Gemini 2.5 Pro
  10. 72.0 o4-mini
  11. 72.0 Claude Opus 4
  12. 71.4 DeepSeek-R1
  13. 64.9 Claude 3.7 Sonnet
  14. 61.7 o1
  15. 61.3 Claude Sonnet 4
  16. 60.4 o3-mini
  17. 59.6 Qwen3-235B-A22B
  18. 59.6 Qwen3-235B-A22B-Instruct
  19. 59.1 Kimi K2
  20. 59.1 Kimi K2
  21. 56.9 DeepSeek-R1
  22. 55.1 DeepSeek-V3
  23. 55.1 Gemini 2.5 Flash
  24. 53.3 Grok 3
  25. 52.4 GPT-4.1
  26. 51.6 Claude 3.5 Sonnet (October 2024)
  27. 49.3 Grok-3 mini
  28. 48.4 DeepSeek-V3
  29. 47.1 Gemini 2.5 Flash
  30. 45.3 GPT-4o
  31. 44.9 GPT-4.5
  32. 41.8 gpt-oss-120b
  33. 40.0 Qwen3-32B
  34. 38.2 Gemini 2.0 Flash
  35. 35.6 Gemini 2.0 Pro
  36. 32.9 o1-mini
  37. 32.4 GPT-4.1 mini
  38. 28.0 Claude 3.5 Haiku
  39. 27.1 GPT-4o
  40. 23.1 GPT-4o
  41. 21.8 Qwen2.5-Max
  42. 20.9 QwQ-32B
  43. 18.2 GPT-4o
  44. 18.2 Gemini 2.0 Flash Thinking
  45. 17.8 DeepSeek-V2.5
  46. 16.4 Qwen2.5-Coder-32B-Instruct
  47. 15.6 Llama 4 Maverick
  48. 12.9 Yi-Lightning
  49. 12.0 Cohere Command A
  50. 11.1 Codestral
  51. 10.2 openhands-lm-32b-v0.1
  52. 8.90 GPT-4.1 nano
  53. 4.90 Gemma 3 27B
  54. 3.60 GPT-4o mini
  1. 1 GPT-5 high 88%
  2. 2 o3-pro high 84.9%
  3. 3 Gemini 2.5 Pro Jun 2025, 32K 83.1%
  4. 4 o3 high 81.3%
  5. 5 Grok 4 79.6%
  6. 6 Gemini 2.5 Pro May 2025 76.9%
  7. 7 DeepSeek-V3.2-Exp thinking 74.2%
  8. 8 DeepSeek-V3.2 Thinking 74.2%
  9. 9 Gemini 2.5 Pro Mar 2025 72.9%
  10. 10 o4-mini high 72%
  11. 11 Claude Opus 4 32K 72%
  12. 12 DeepSeek-R1 May 2025 71.4%
  13. 13 Claude 3.7 Sonnet 32k thinking 64.9%
  14. 14 o1 high 61.7%
  15. 15 Claude Sonnet 4 32K 61.3%
  16. 16 o3-mini high 60.4%
  17. 17 Qwen3-235B-A22B 59.6%
  18. 18 Qwen3-235B-A22B-Instruct Jul 2025 59.6%
  19. 19 Kimi K2 Jul 2025 59.1%
  20. 20 Kimi K2 Sep 2025 59.1%
  21. 21 DeepSeek-R1 56.9%
  22. 22 DeepSeek-V3 Mar 2025 55.1%
  23. 23 Gemini 2.5 Flash May 2025, 23K 55.1%
  24. 24 Grok 3 53.3%
  25. 25 GPT-4.1 52.4%
  26. 26 Claude 3.5 Sonnet (October 2024) 51.6%
  27. 27 Grok-3 mini high 49.3%
  28. 28 DeepSeek-V3 48.4%
  29. 29 Gemini 2.5 Flash Apr 2025 47.1%
  30. 30 GPT-4o Mar 2025 45.3%
  31. 31 GPT-4.5 44.9%
  32. 32 gpt-oss-120b high 41.8%
  33. 33 Qwen3-32B 40%
  34. 34 Gemini 2.0 Flash Dec 2024 38.2%
  35. 35 Gemini 2.0 Pro 35.6%
  36. 36 o1-mini unknown thinking 32.9%
  37. 37 GPT-4.1 mini 32.4%
  38. 38 Claude 3.5 Haiku Oct 2024 28%
  39. 39 GPT-4o Jan 2025 27.1%
  40. 40 GPT-4o Aug 2024 23.1%
  41. 41 Qwen2.5-Max 21.8%
  42. 42 QwQ-32B 20.9%
  43. 43 GPT-4o Nov 2024 18.2%
  44. 44 Gemini 2.0 Flash Thinking Jan 2025 18.2%
  45. 45 DeepSeek-V2.5 Sep 2024 17.8%
  46. 46 Qwen2.5-Coder-32B-Instruct 16.4%
  47. 47 Llama 4 Maverick 15.6%
  48. 48 Yi-Lightning 12.9%
  49. 49 Cohere Command A 12%
  50. 50 Codestral 11.1%
  51. 51 openhands-lm-32b-v0.1 10.2%
  52. 52 GPT-4.1 nano 8.90%
  53. 53 Gemma 3 27B 4.90%
  54. 54 GPT-4o mini 3.60%
#ModelScoreSettingDateSource
1 GPT-5 88% high Aug 2025 Epoch AI
2 o3-pro 84.9% high Jun 2025 Epoch AI
3 Gemini 2.5 Pro Jun 2025 83.1% 32K Jun 2025 Epoch AI
4 o3 81.3% high Apr 2025 Epoch AI
5 Grok 4 79.6% Jul 2025 Epoch AI
6 Gemini 2.5 Pro May 2025 76.9% May 2025 Epoch AI
7 DeepSeek-V3.2-Exp 74.2% thinking Sep 2025 Epoch AI
8 DeepSeek-V3.2 74.2% Thinking Dec 2025 Epoch AI
9 Gemini 2.5 Pro Mar 2025 72.9% Mar 2025 Epoch AI
10 o4-mini 72% high Apr 2025 Epoch AI
11 Claude Opus 4 72% 32K May 2025 Epoch AI
12 DeepSeek-R1 May 2025 71.4% May 2025 Epoch AI
13 Claude 3.7 Sonnet 64.9% 32k thinking Feb 2025 Epoch AI
14 o1 61.7% high Dec 2024 Epoch AI
15 Claude Sonnet 4 61.3% 32K May 2025 Epoch AI
16 o3-mini 60.4% high Jan 2025 Epoch AI
17 Qwen3-235B-A22B 59.6% Apr 2025 Epoch AI
18 Qwen3-235B-A22B-Instruct Jul 2025 59.6% Jul 2025 Epoch AI
19 Kimi K2 Jul 2025 59.1% Jul 2025 Epoch AI
20 Kimi K2 Sep 2025 59.1% Sep 2025 Epoch AI
21 DeepSeek-R1 56.9% Jan 2025 Epoch AI
22 DeepSeek-V3 Mar 2025 55.1% Mar 2025 Epoch AI
23 Gemini 2.5 Flash May 2025 55.1% 23K May 2025 Epoch AI
24 Grok 3 53.3% Apr 2025 Epoch AI
25 GPT-4.1 52.4% Apr 2025 Epoch AI
26 Claude 3.5 Sonnet (October 2024) 51.6% Oct 2024 Epoch AI
27 Grok-3 mini 49.3% high Apr 2025 Epoch AI
28 DeepSeek-V3 48.4% Dec 2024 Epoch AI
29 Gemini 2.5 Flash Apr 2025 47.1% Apr 2025 Epoch AI
30 GPT-4o Mar 2025 45.3% Mar 2025 Epoch AI
31 GPT-4.5 44.9% Feb 2025 Epoch AI
32 gpt-oss-120b 41.8% high Aug 2025 Epoch AI
33 Qwen3-32B 40% Apr 2025 Epoch AI
34 Gemini 2.0 Flash Dec 2024 38.2% Dec 2024 Epoch AI
35 Gemini 2.0 Pro 35.6% Feb 2025 Epoch AI
36 o1-mini 32.9% unknown thinking Sep 2024 Epoch AI
37 GPT-4.1 mini 32.4% Apr 2025 Epoch AI
38 Claude 3.5 Haiku 28% Oct 2024 Oct 2024 Epoch AI
39 GPT-4o Jan 2025 27.1% Jan 2025 Epoch AI
40 GPT-4o Aug 2024 23.1% Aug 2024 Epoch AI
41 Qwen2.5-Max 21.8% Jan 2025 Epoch AI
42 QwQ-32B 20.9% Mar 2025 Epoch AI
43 GPT-4o Nov 2024 18.2% Nov 2024 Epoch AI
44 Gemini 2.0 Flash Thinking Jan 2025 18.2% Jan 2025 Epoch AI
45 DeepSeek-V2.5 Sep 2024 17.8% Sep 2024 Epoch AI
46 Qwen2.5-Coder-32B-Instruct 16.4% Nov 2024 Epoch AI
47 Llama 4 Maverick 15.6% Apr 2025 Epoch AI
48 Yi-Lightning 12.9% Dec 2024 Epoch AI
49 Cohere Command A 12% Mar 2025 Epoch AI
50 Codestral 11.1% Jan 2025 Epoch AI
51 openhands-lm-32b-v0.1 10.2% Mar 2024 Epoch AI
52 GPT-4.1 nano 8.90% Apr 2025 Epoch AI
53 Gemma 3 27B 4.90% Mar 2025 Epoch AI
54 GPT-4o mini 3.60% Jul 2024 Epoch AI

The benchmark's own leaderboard has 58 entries, 46 of them models we track. Labs report scores for 29 models in their cards. We plot every source in the charts below.

More effort adds a median 7 points

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
GPT-5 aider, diff edits
low high
88%
o3 aider, diff edits
high
81.3%
Claude Opus 4 aider, diff edits
32k thinking
72%
Claude 3.7 Sonnet aider, diff edits
no thinking
64.9%
Claude Sonnet 4 aider, diff edits
no thinking
61.3%
o3-mini aider, diff edits
medium high
60.4%
Grok-3 mini aider, whole edits
low high
49.3%
ModelEffortHarnessScoreCost per taskOutput tokens
GPT-5 high aider, diff edits 88%
GPT-5 medium aider, diff edits 86.7%
o3-pro high aider, diff edits 84.9%
Gemini 2.5 Pro Jun 2025 aider, diff-fenced edits 83.1%
o3 high aider, diff edits 81.3%
GPT-5 low aider, diff edits 81.3%
Grok 4 high aider, diff edits 79.6%
Gemini 2.5 Pro Jun 2025 aider, diff-fenced edits 79.1%
o3 aider, diff edits 76.9%
Gemini 2.5 Pro May 2025 aider, diff-fenced edits 76.9%
Gemini 2.5 Pro Mar 2025 aider, diff-fenced edits 72.9%
o4-mini high aider, diff edits 72%
Claude Opus 4 32k thinking aider, diff edits 72%
DeepSeek-R1 May 2025 aider, diff edits 71.4%
Claude Opus 4 aider, diff edits 70.7%
Claude 3.7 Sonnet 32k thinking tokens aider, diff edits 64.9%
o1 high aider, diff edits 61.7%
Claude Sonnet 4 32k thinking aider, diff edits 61.3%
o3-mini high aider, diff edits 60.4%
Claude 3.7 Sonnet no thinking aider, diff edits 60.4%
Kimi K2 Jul 2025 aider, diff edits 59.1%
DeepSeek-R1 aider, diff edits 56.9%
Claude Sonnet 4 no thinking aider, diff edits 56.4%
DeepSeek-V3 Mar 2025 aider, diff edits 55.1%
Gemini 2.5 Flash May 2025 aider, diff edits 55.1%
o3-mini medium aider, diff edits 53.8%
Grok 3 aider, diff edits 53.3%
GPT-4.1 aider, diff edits 52.4%
Claude 3.5 Sonnet aider, diff edits 51.6%
Grok-3 mini high aider, whole edits 49.3%
Gemini 2.5 Flash Apr 2025 aider, diff edits 47.1%
GPT-4o Mar 2025 aider, diff edits 45.3%
GPT-4.5 aider, diff edits 44.9%
Gemini 2.5 Flash May 2025 aider, diff edits 44%
gpt-oss-120b high aider, diff edits 41.8%
Qwen3-32B aider, diff edits 40%
Gemini 2.0 Pro aider, whole edits 35.6%
Grok-3 mini low aider, whole edits 34.7%
o1-mini aider, whole edits 32.9%
GPT-4.1 mini aider, diff edits 32.4%
Claude 3.5 Haiku aider, diff edits 28%
GPT-4o Jan 2025 aider, diff edits 27.1%
GPT-4o Aug 2024 aider, diff edits 23.1%
Gemini 2.0 Flash Dec 2024 aider, whole edits 22.2%
QwQ-32B aider, diff edits 20.9%
GPT-4o Nov 2024 aider, diff edits 18.2%
Gemini 2.0 Flash Thinking Jan 2025 aider, diff edits 18.2%
Qwen2.5-Coder-32B-Instruct aider, whole edits 16.4%
Llama 4 Maverick aider, whole edits 15.6%
Yi-Lightning aider, whole edits 12.9%
Cohere Command A aider, whole edits 12%
Codestral aider, whole edits 11.1%
openhands-lm-32b-v0.1 aider, whole edits 10.2%
GPT-4.1 nano aider, whole edits 8.90%
Qwen2.5-Coder-32B-Instruct aider, diff edits 8.00%
Gemma 3 27B aider, whole edits 4.90%
GPT-4o mini aider, whole edits 3.60%

8 models have held number one since its release in Dec 2024

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, Epoch AI’s runs and what labs claim in their cards.

Anthropic OpenAI Google DeepMind DeepSeek
Own leaderboard Passed 85% in Aug 2025, by GPT-5
GPT-5 88%
Epoch AI runs Passed 85% in Aug 2025, by GPT-5
GPT-5 88%
Lab claims Passed 85% in Aug 2025, by GPT-5
GPT-5 88%

The best score passed 85% in Aug 2025

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 Run or collected by Epoch AI Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2024 2025 2026 Near the ceiling (85%) GPT-5 Gemini 2.5 Pro Gemini 2.5 Pro Claude 3.7 Sonnet Claude 3.5 Sonnet openhands-lm-32b-v0.1 0% 20% 40% 60% 80% 100% 2024 2025 2026 Near the ceiling (85%) GPT-5 o3

Labs headline it less often now, last in Feb 2026

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.

DeepSeek
Google DeepMind
ByteDance Seed
OpenAI
Qwen
Mistral AI
Moonshot AI
DateLaunch postScoreAs the post writes it
14 Feb 2026 Seed2.0 Pro 80 Aider Polyglot
2 Feb 2026 qwen3-coder-next 66.2 Aider
29 Sep 2025 DeepSeek-V3.2-Exp 74.5 Aider-Polyglot
22 Sep 2025 DeepSeek-V3.1-Terminus 76.1 Aider-Polyglot
21 Aug 2025 DeepSeek-V3.1 76.3 Aider-Polyglot (Acc.)
7 Aug 2025 GPT-5 nano 48.4 Aider polyglot (diff)
7 Aug 2025 GPT-5 mini 71.6 Aider polyglot (diff)
7 Aug 2025 GPT-5 88 Aider Polyglot Multi-language code editing
22 Jul 2025 Qwen3-Coder-480B-A35B 61.8 Aider-Polyglot
11 Jul 2025 Kimi K2 60 Aider-Polyglot, Acc
17 Jun 2025 Gemini 2.5 Flash 61.9 Aider Polyglot whole 61.9 / diff 56.7
17 Jun 2025 Gemini 2.5 Flash-Lite 27.1 Aider Polyglot (whole), Thinking
17 Jun 2025 Gemini 2.5 Pro 82.2 Aider Polyglot (diff-fenced)
10 Jun 2025 Magistral Medium 1.0 47.1 Aider-Polyglot
28 May 2025 DeepSeek-R1 71.6 Aider-Polyglot (Acc.)
29 Apr 2025 Qwen3-235B-A22B Aider (Pass@2)
17 Apr 2025 Gemini 2.5 Flash 51.1 Aider Polyglot whole 51.1 / diff-fenced 44.2
16 Apr 2025 o4-mini 68.9 Aider Polyglot whole 68.9 (diff 58.2), o4-mini-high
16 Apr 2025 o3 81.3 Aider Polyglot Code Editing, o3-high (whole)
14 Apr 2025 GPT-4.1 nano 9.8 Aider's polyglot: whole 9.8 (diff 6.2)
14 Apr 2025 GPT-4.1 mini 34.7 Aider's polyglot: whole 34.7 (diff 31.6)
14 Apr 2025 GPT-4.1 52.9 Aider's polyglot: diff
10 Apr 2025 Seed-Thinking-v1.5 54.2 Aider Polyglot
25 Mar 2025 Gemini 2.5 Pro 74 Aider Polyglot, whole (diff = 68.6%)
20 Jan 2025 DeepSeek-R1 53.3 Aider-Polyglot (Acc.)
26 Dec 2024 DeepSeek-V3 49.6 Aider-Polyglot (Acc.)

17 system cards report it, from 9 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
28 Feb 2026 qwen3-coder-next technical report 66.2 Aider-Polyglot (Table 7) Capabilities
14 Feb 2026 Seed2.0 Pro model card 80 Aider Polyglot (Table 11) Capabilities
30 Sep 2025 Code World Model (CWM) technical report 35.1 Aider Polyglot, whole file format, pass 1@2 (Table 6) Capabilities
29 Sep 2025 DeepSeek-V3.2-Exp technical report 74.5 Aider-Polyglot (Acc.), DeepSeek-V3.2-Exp (Table 1) Capabilities
25 Sep 2025 gemini-robotics-er-1.5-preview technical report 57.3 Aider Polyglot (Table 20, thinking on). Thinking off 44.4 Capabilities
5 Aug 2025 gpt-oss-20b model card 34.2 Aider Polyglot, high reasoning (Table 3) Capabilities
5 Aug 2025 gpt-oss-120b model card 44.4 Aider Polyglot, high reasoning (Table 3) Capabilities
21 Jul 2025 Kimi K2 technical report 60 Aider-Polyglot (Acc.) (Table 3) Capabilities
7 Jul 2025 Gemini 2.5 Pro technical report 82.2 Aider Polyglot, diff or diff-fenced format, average of 3 trials (Table 3) Capabilities
7 Jul 2025 Gemini 2.5 Flash technical report 56.7 Aider Polyglot, diff or diff-fenced format, average of 3 trials (Table 3) Capabilities
12 Jun 2025 Magistral Medium 1.0 technical report 47.1 Aider Polyglot (Table 2) Capabilities
29 Apr 2025 Gemini 2.5 Flash model card 51.1 Aider Polyglot, whole 51.1, diff 44.2 (Table, p.4) Capabilities
28 Apr 2025 Gemini 2.5 Pro model card 74 Aider Polyglot, whole edit format; diff format = 68.6% (Table, p.4, column Gemini 2.5 Pro Experimental 03-25) Capabilities
10 Apr 2025 Seed-Thinking-v1.5 technical report 54.2 Aider Polyglot (Table 2) Capabilities
1 Apr 2025 Cohere Command A technical report 14.7 Aider Polyglot (Table 13) Capabilities
20 Jan 2025 DeepSeek-R1 technical report 53.3 Aider-Polyglot (Acc.), DeepSeek-R1 (Table 8, same as final column of Table 3) Capabilities
26 Dec 2024 DeepSeek-V3 technical report 49.6 Aider-Polyglot (Acc.), DeepSeek-V3 (Table 6) Capabilities

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.

OpenAI Anthropic Google DeepMind
Aider Polyglot Passed 85% in Aug 2025, by GPT-5
GPT-5 88%
Aider-Edit Not tested since Oct 2025
Claude 3.5 Sonnet 84.2%
Aider Polyglot Passed 85% in Aug 2025, by GPT-5
GPT-5 88%
Replaces Aider-Edit

More in Coding

All 243 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%
  • SWE-bench Multilingual In use 300 GitHub-issue tasks across nine programming languages in SWE-bench format. Patch graded by repository tests. Claude Opus 5.5 93.9%
  • 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%
  • 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