SpireClimb
Integrity rank #29 of 65 · 7 models scored · top score 25.67 · GPT-6 Astra
unknown
Strongest on contamination resistance, weakest on discrimination. 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-6 Astra | 25.67 | unverifiedT2 | 2026-09-03 |
| 2 | GPT-6.1 Sol | 24.67 | unverifiedT2 | 2026-09-29 |
| 3 | GPT-6 Sol | 7.5 | unverifiedT2 | 2026-09-22 |
| 4 | Claude Opus 5.5 | 3.33 | unverifiedT2 | 2026-09-22 |
| 5 | Claude Fable 5.1 | 2 | unverifiedT2 | 2026-09-01 |
| 6 | Claude Sonnet 5.5 | 1.33 | unverifiedT2 | 2026-09-28 |
| 7 | GPT-6 Luna | 0 | unverifiedT2 | 2026-09-22 |
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.
Seven models are scored on this benchmark, and the best-to-worst score spread is 25.67 points. A spread this wide supports real separation between models, so rank differences are more likely to reflect genuine capability differences than scoring noise.
3 cited facts
The benchmark's Benchmark Integrity Index is 87, grade A, ranking 29th among 65 benchmarks. Its weakest integrity component is discrimination, which narrows how sharply the benchmark can separate frontier models under its disclosed harness. This score reflects the benchmark's health as a discriminator of frontier models under a disclosed harness, not any model's capability.
6 cited facts
This benchmark is not saturated: its top score is 25.67, leaving 74.33 points of headroom to the ceiling. Only 2 models cluster within a couple of points of the top, so the leading group is small. Because it is not saturated, there is still genuine room to separate the field at the top, so small differences near the top can reflect real capability gaps rather than noise.
5 cited facts
Harness comparability is reported as consistent harness, so scores are intended to be directly comparable within this disclosed evaluation setup rather than across differing harnesses. Test set privacy is unknown, so it cannot be determined whether the evaluation data are public or held out; the usual extra contamination scrutiny for public test sets therefore cannot be applied or ruled out here. Contamination history is unknown and the harness notes describe metadata and carried-forward optimized-scaffold flags, so scores should be read as task performance under the disclosed harness, not as deployed capability.
4 cited facts
This benchmark's intended target is not documented, so what it actually measures cannot be stated from the available record. The record likewise does not describe how the test was built, leaving its construction procedure undocumented. Its operating assumptions are also undocumented, so there is no stated basis for the conditions under which results should be interpreted. The sharpest caveat is that no limitations and no human reference point are documented, so a reader cannot tell what the benchmark fails to capture or where its results are most likely to mislead, and what it measures being unknown means even the quantity being scored is unclear.
6 cited facts
SpireClimb: SpireClimb as reported in Epoch AI's Capabilities Index CSV.
GPT-6 Astra leads SpireClimb at 25.67. The full leaderboard above lists every recorded measurement, not just the headline number.
7 models have recorded scores on SpireClimb, spanning a score spread of 25.67.
tensor.news grades SpireClimb A for integrity (score 87/100), ranking #29 of 65 benchmarks we assess across discrimination, saturation, contamination resistance, harness comparability, and freshness.
No — SpireClimb 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 SpireClimb are reasonably apples-to-apples.
Every SpireClimb measurement is source-backed and tagged reproduced, self-reported, or unverified; the underlying record cites epoch.ai/data/processed_data_for_eci.csv.