Aider polyglot
Integrity rank #32 of 65 · 45 models scored · top score 88 · GPT-5
unknown
Strongest on discrimination, weakest on contamination resistance. Caveat: the public test set does not rule out training-data contamination.
Reported scores — protocols may differ. How we compare
| # | Model | Score | Evidence | Measured |
|---|---|---|---|---|
| 1 | GPT-5 | 88 | unverified· optimizedT1 | 2025-08-07 |
| 2 | o3-pro | 84.9 | unverified· optimizedT1 | 2025-06-10 |
| 3 | Gemini 2.5 Pro (Jun 2025) | 83.1 | unverified· optimizedT1 | 2025-06-05 |
| 4 | o3 | 81.3 | unverified· optimizedT1 | 2025-04-16 |
| 5 | Grok 4 | 79.6 | unverified· optimizedT1 | 2025-07-09 |
| 6 | Gemini 2.5 Pro (May 2025) | 76.9 | unverified· optimizedT1 | 2025-05-06 |
| 7 | DeepSeek-V3.2 | 74.2 | unverified· optimizedT1 | 2025-12-01 |
| 8 | DeepSeek-V3.2-Exp | 74.2 | unverified· optimizedT1 | 2025-09-29 |
| 9 | Gemini 2.5 Pro (Mar 2025) | 72.9 | unverified· optimizedT1 | 2025-03-31 |
| 10 | Claude Opus 4 | 72 | unverified· optimizedT1 | 2025-05-22 |
| 11 | o4-mini | 72 | unverified· optimizedT1 | 2025-04-16 |
| 12 | DeepSeek-R1 (May 2025) | 71.4 | unverified· optimizedT1 | 2025-05-28 |
| 13 | Claude 3.7 Sonnet | 64.9 | unverified· optimizedT1 | 2025-02-24 |
| 14 | o1 | 61.7 | unverified· optimizedT1 | 2024-12-17 |
| 15 | Claude Sonnet 4 | 61.3 | unverified· optimizedT1 | 2025-05-22 |
| 16 | o3-mini | 60.4 | unverified· optimizedT1 | 2025-01-31 |
| 17 | Qwen3-235B-A22B | 59.6 | unverified· optimizedT1 | 2025-04-28 |
| 18 | Qwen3-235B-A22B-Instruct (Jul 2025) | 59.6 | unverified· optimizedT1 | 2025-07-25 |
| 19 | Kimi K2 (Jul 2025) | 59.1 | unverified· optimizedT1 | 2025-07-12 |
| 20 | DeepSeek-R1 | 56.9 | unverified· optimizedT1 | 2025-01-20 |
| 21 | Gemini 2.5 Flash (May 2025) | 55.1 | unverified· optimizedT1 | 2025-05-20 |
| 22 | DeepSeek-V3 (Mar 2025) | 55.1 | unverified· optimizedT1 | 2025-03-24 |
| 23 | Grok 3 | 53.3 | unverified· optimizedT1 | 2025-04-09 |
| 24 | GPT-4.1 | 52.4 | unverified· optimizedT1 | 2025-04-14 |
| 25 | Claude 3.5 Sonnet (October 2024) | 51.6 | unverified· optimizedT1 | 2024-10-22 |
| 26 | Grok-3 mini | 49.3 | unverified· optimizedT1 | 2025-04-09 |
| 27 | DeepSeek-V3 | 48.4 | unverified· optimizedT1 | 2024-12-26 |
| 28 | Gemini 2.5 Flash (Apr 2025) | 47.1 | unverified· optimizedT1 | 2025-04-17 |
| 29 | GPT-4.5 | 44.9 | unverified· optimizedT1 | 2025-02-27 |
| 30 | gpt-oss-120b | 41.8 | unverified· optimizedT1 | 2025-08-05 |
| 31 | Qwen3-32B | 40 | unverifiedT1 | 2025-04-28 |
| 32 | Gemini 2.0 Flash (Dec 2024) | 38.2 | unverified· optimizedT1 | 2024-12-11 |
| 33 | Gemini 2.0 Pro | 35.6 | unverified· optimizedT1 | 2025-02-05 |
| 34 | o1-mini | 32.9 | unverified· optimizedT1 | 2024-09-12 |
| 35 | GPT-4.1 mini | 32.4 | unverified· optimizedT1 | 2025-04-14 |
| 36 | Claude 3.5 Haiku | 28 | unverified· optimizedT1 | 2024-10-22 |
| 37 | GPT-4o (Aug 2024) | 23.1 | unverified· optimizedT1 | 2024-08-06 |
| 38 | Qwen2.5-Max | 21.8 | unverified· optimizedT1 | 2025-01-25 |
| 39 | QwQ-32B | 20.9 | unverifiedT1 | 2025-03-05 |
| 40 | Gemini 2.0 Flash Thinking (Jan 2025) | 18.2 | unverified· optimizedT1 | 2025-01-21 |
| 41 | GPT-4o (Nov 2024) | 18.2 | unverified· optimizedT1 | 2024-11-20 |
| 42 | Llama 4 Maverick | 15.6 | unverified· optimizedT1 | 2025-04-05 |
| 43 | GPT-4.1 nano | 8.9 | unverified· optimizedT1 | 2025-04-14 |
| 44 | Gemma 3 27B | 4.9 | unverified· optimizedT1 | 2025-03-12 |
| 45 | GPT-4o mini | 3.6 | unverified· optimizedT1 | 2024-07-18 |
Continue from this leaderboard into the tightest head-to-head reads.
Weighted composite. Each dimension opens to the source-backed sub-signals it is computed from.
Claims drawn from cited facts, not live model generation.
This benchmark scores 45 models, and the gap between the best and worst scores spans 84.4 points. That wide spread supports real separation between models, so ranks near each other are still likely to reflect genuine capability differences rather than noise.
3 cited facts
This benchmark posts an integrity score of 86, a grade of A, ranking 32nd among the 62 benchmarks in the tracked universe — a measure of its health as a discriminator of frontier models under a disclosed harness, not of any model's capability. The single weakest component of the integrity breakdown is contamination, largely because the test-set privacy is unknown and a sizable share of the contamination window overlaps training data; this narrows how much the reader should trust the very top scores, since strong results may reflect familiarity with the items rather than genuine frontier discrimination.
6 cited facts
This benchmark is not saturated: the leading result stands at 88.0 points, leaving twelve points of headroom before the ceiling is reached. Only one model sits within a couple of points of that top score, so the top of the leaderboard is not yet crowded. Because the benchmark is not saturated, there is still genuine room to separate the field at the top, and differences among the best models can be read as real capability gaps rather than noise.
6 cited facts
For this benchmark, the harness is consistent, so scores are directly comparable within that harness, while test-set privacy is unknown. In practice, low comparability would mean scores from different harnesses should not be compared head-to-head, and a public test set would mean high scores deserve more scrutiny for possible contamination than a held-out one; because privacy is unknown, that contamination-scrutiny judgment cannot be resolved here. Contamination history is unknown, and the harness notes list model, version, performance, optimized flag, score source, and dates, which suggests some harness sensitivity; results should therefore be read as task performance under a disclosed harness rather than as deployed capability.
7 cited facts
This benchmark's design cannot be described in plain terms here, because the field stating what it measures is not documented. Likewise, how the test was built is not documented, so the construction procedure is unknown. The assumptions the benchmark rests on are also not documented, and no human baseline is recorded for comparison. The sharpest caveat a reader should hold onto is that the limitations are not documented either, so there is no stated guidance on what this benchmark cannot tell you or where it is most likely to mislead.
8 cited facts
Aider polyglot: Aider polyglot as reported in Epoch AI's Capabilities Index CSV.
GPT-5 leads Aider polyglot at 88. The full leaderboard above lists every recorded measurement, not just the headline number.
45 models have recorded scores on Aider polyglot, spanning a score spread of 84.4.
tensor.news grades Aider polyglot A for integrity (score 85/100), ranking #32 of 65 benchmarks we assess across discrimination, saturation, contamination resistance, harness comparability, and freshness.
No — Aider polyglot still has headroom and continues to discriminate between models rather than bunching them at the ceiling.
Its test set is unknown, so contamination risk is on the table: a high score may partly reflect training-data overlap rather than capability.
Scores are reported under a consistent harness, so comparisons on Aider polyglot are reasonably apples-to-apples.
Every Aider polyglot measurement is source-backed and tagged reproduced, self-reported, or unverified; the underlying record cites epoch.ai/data/eci_benchmarks.csv.