MirrorCode
Integrity rank #19 of 61 · 6 models scored · top score 63.89 · Claude Fable 5
unknown
Strongest on discrimination, weakest on saturation headroom. 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 Fable 5 | 63.89 | reproducedT1 | 2026-06-09 |
| 2 | Claude Opus 4.7 | 31.11 | reproducedT1 | 2026-04-16 |
| 3 | GPT-5.6 Sol | 20 | reproducedT1 | 2026-07-09 |
| 4 | GPT-5.4 | 15.56 | reproducedT1 | 2026-03-05 |
| 5 | GPT-5.5 | 10 | reproducedT1 | 2026-04-23 |
| 6 | 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 evaluates 6 models, with scores spanning 55.0 points from best to worst. Such a wide spread indicates that the observed rank differences reflect genuine separation between models rather than noise.
2 cited facts
The benchmark's integrity score is 97 (grade A), placing it 19th out of 61 benchmarks in the integrity rank. This reflects strong health as a discriminator of frontier models under a disclosed harness, with the weakest component being saturation, meaning the ceiling is crowded and top scores become harder to distinguish.
4 cited facts
This benchmark is not saturated, evidenced by the top score standing at 63.89 and leaving 36.11 points of headroom to the ceiling. Only one model clusters within a couple of points of the top, so the ranking is not yet compressed into noise. For the reader, this means there is still genuine room to separate the field at the top, and small differences among leading models can carry real signal.
5 cited facts
The disclosed harness is consistent, so task performance scores are directly comparable across reported runs. The privacy status of this benchmark's test set is unknown, and its contamination history is also unknown, so the degree to which scores might reflect memorization cannot be assessed from these facts. Harness sensitivity is qualitatively limited to metadata tracking—recorded model, version, performance, optimization flag, score source, and release/model dates—so scores should be interpreted as task performance under this disclosed harness, not deployed capability.
4 cited facts
This benchmark's intended measurement is not documented in the provided facts, and its construction is likewise unspecified. The underlying assumptions and limitations are also undocumented, so any inferences about what this benchmark can or cannot tell a reader would be speculative. A human baseline is not documented, so no comparative judgment against human performance is available from this record.
5 cited facts
MirrorCode: MirrorCode as reported in Epoch AI's Capabilities Index CSV.
Claude Fable 5 leads MirrorCode at 63.89 — independently reproduced. The full leaderboard above lists every recorded measurement, not just the headline number.
6 models have recorded scores on MirrorCode, spanning a score spread of 55.
tensor.news grades MirrorCode A for integrity (score 97/100), ranking #19 of 61 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.