MirrorCode
Integrity rank #20 of 65 · 9 models scored · top score 77.4 · Claude Opus 5.5
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 | Claude Opus 5.5 | 77.4 | unverifiedT2 | 2026-09-22 |
| 2 | Claude Fable 5.1 | 73.3 | unverifiedT1 | 2026-09-01 |
| 3 | Claude Fable 5 | 63.89 | reproducedT1 | 2026-06-09 |
| 4 | GPT-6 Astra | 46.7 | unverifiedT2 | 2026-09-03 |
| 5 | Claude Opus 4.7 | 31.11 | reproducedT1 | 2026-04-16 |
| 6 | GPT-5.6 Sol | 20 | reproducedT1 | 2026-07-09 |
| 7 | GPT-5.4 | 15.56 | reproducedT1 | 2026-03-05 |
| 8 | GPT-5.5 | 10 | reproducedT1 | 2026-04-23 |
| 9 | Gemini 3.1 Pro | 8.89 | reproducedT1 | 2026-02-19 |
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 9 models, and the spread between the best and worst scores is 68.51 points. That wide spread supports real separation between models, whereas a narrow spread would mean close scores may not reflect a genuine capability gap.
3 cited facts
Its Benchmark Integrity Index score is 91, with grade A and rank 20 out of 65 tracked benchmarks.
4 cited facts
This benchmark is not saturated: the leading score stands at 77.4 points, leaving 22.6 points of headroom to the ceiling, and only one model sits within a couple of points of the top. Because it is not saturated, there is still genuine room to separate the field at the top, so differences among the strongest models can reflect real capability gaps rather than noise.
5 cited facts
For this benchmark, scores are produced under a consistent harness, so they are directly comparable to other scores generated with that same disclosed harness and should be read as task performance under that harness. The test set privacy is unknown, so it is not established whether the evaluation data are public or held out; if the set were public, unusually high scores would warrant more contamination scrutiny than a held-out set, and contamination history is likewise unknown. Because harness sensitivity is not documented and the harness notes only describe reported metadata, comparisons across different harnesses should not be made head-to-head, and results should not be framed as deployed capability.
5 cited facts
This benchmark's design cannot be described in plain terms from the available record, because the specific capability it measures is not documented. Equally, the procedure used to build the test and the assumptions it rests on are not documented, so a reader cannot say how items were assembled or what the benchmark takes for granted. The sharpest caveat follows directly: with no recorded limitations and no documented human baseline, there is no stated boundary on what this benchmark can and cannot tell you, and no reference point for interpreting a score.
5 cited facts
MirrorCode: MirrorCode as reported in Epoch AI's Capabilities Index CSV.
Claude Opus 5.5 leads MirrorCode at 77.4. The full leaderboard above lists every recorded measurement, not just the headline number.
9 models have recorded scores on MirrorCode, spanning a score spread of 68.51.
tensor.news grades MirrorCode A for integrity (score 91/100), ranking #20 of 65 benchmarks we assess across discrimination, saturation, contamination resistance, harness comparability, and freshness.
No — MirrorCode 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 MirrorCode are reasonably apples-to-apples.
Every MirrorCode measurement is source-backed and tagged reproduced, self-reported, or unverified; the underlying record cites epoch.ai/data/eci_benchmarks.csv.