FrontierCode
Integrity rank #7 of 61 · 25 models scored · top score 53.5 · 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 | 53.5 | unverifiedT1 | 2026-06-09 |
| 2 | Claude Opus 5 | 53.4 | unverifiedT1 | 2026-07-24 |
| 3 | Grok 4.6 | 48.01 | unverifiedT1 | 2026-08-12 |
| 4 | GPT-5.6 Sol | 47.5 | unverifiedT1 | 2026-07-09 |
| 5 | Claude Opus 4.8 | 46.5 | unverifiedT1 | 2026-05-28 |
| 6 | Kimi K3 | 44.2 | unverifiedT1 | 2026-07-16 |
| 7 | Gemini 3.7 Flash | 43.59 | unverifiedT1 | 2026-08-13 |
| 8 | GPT-5.5 | 43 | unverifiedT1 | 2026-04-23 |
| 9 | Claude Sonnet 5 | 42.7 | unverifiedT1 | 2026-06-30 |
| 10 | Grok 4.5 | 42.4 | unverifiedT1 | 2026-07-08 |
| 11 | GPT-5.6 Terra | 41.3 | unverifiedT1 | 2026-07-09 |
| 12 | GPT-5.6 Luna | 39.8 | unverifiedT1 | 2026-07-09 |
| 13 | Claude Opus 4.7 | 38.5 | unverifiedT1 | 2026-04-16 |
| 14 | Gemini 3.6 Flash | 34.37 | unverifiedT1 | 2026-07-21 |
| 15 | Kimi K2.7 Code | 30.1 | unverifiedT1 | 2026-06-12 |
| 16 | GPT-5.4 Mini | 27.04 | unverifiedT1 | 2026-03-17 |
| 17 | Claude Opus 4.6 | 26.64 | unverifiedT1 | 2026-02-05 |
| 18 | GLM-5.2 | 24.5 | unverifiedT1 | 2026-06-16 |
| 19 | Claude Sonnet 4.6 | 24.31 | unverifiedT1 | 2026-02-17 |
| 20 | DeepSeek V4 Flash 0731 | 18.8 | unverifiedT1 | 2026-07-31 |
| 21 | DeepSeek-V4-Pro | 17.6 | unverifiedT1 | 2026-04-24 |
| 22 | MiniMax-M3 | 14.7 | unverifiedT1 | 2026-06-01 |
| 23 | Inkling | 14 | unverifiedT1 | 2026-07-15 |
| 24 | Qwen3.7-Plus | 10.2 | unverifiedT1 | 2026-06-02 |
| 25 | Mistral Medium 3.5 | 8 | unverifiedT1 | 2026-04-29 |
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 25 models, and the gap between the highest and lowest score is 45.5 points. A spread of this size implies that rank differences are likely trustworthy, indicating real separation between model capabilities.
3 cited facts
The benchmark's integrity score is 99, corresponding to grade A, and it ranks 7th among a universe of 61 benchmarks. Saturation is the single weakest component of the integrity breakdown. Saturation means the ceiling is crowded, which limits how much top scores can differentiate among frontier models.
5 cited facts
The benchmark is not saturated, so there is still genuine room to separate models at the top. The top score is 53.5, leaving 46.5 points of headroom to the ceiling. Only 2 models cluster near the top, and they sit within a couple of points of each other, meaning small differences among the leaders should not yet be read as real capability gaps.
5 cited facts
Scores are directly comparable across runs because a consistent harness was used, permitting head-to-head comparison of task performance. The privacy of the test set is unknown; if it were public, high scores would merit additional scrutiny for possible contamination relative to a held-out set. Contamination history is unknown, so no qualitative assessment of prior exposure is possible, and these scores should be read as task performance under the disclosed harness rather than as deployed capability.
4 cited facts
This benchmark's design is not documented: what it measures is unknown, so the intended construct is unspecified. Construction and key assumptions are likewise not documented, meaning the test's assembly and premises cannot be examined. The sharpest caveat is that limitations are not documented, so readers cannot know where this benchmark is most likely to mislead or what it fails to capture.
4 cited facts
FrontierCode: FrontierCode as reported in Epoch AI's Capabilities Index CSV.
Claude Fable 5 leads FrontierCode at 53.5. The full leaderboard above lists every recorded measurement, not just the headline number.
25 models have recorded scores on FrontierCode, spanning a score spread of 45.5.
tensor.news grades FrontierCode A for integrity (score 99/100), ranking #7 of 61 benchmarks we assess across discrimination, saturation, contamination resistance, harness comparability, and freshness.
No — FrontierCode 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 FrontierCode are reasonably apples-to-apples.
Every FrontierCode measurement is source-backed and tagged reproduced, self-reported, or unverified; the underlying record cites epoch.ai/data/eci_benchmarks.csv.