Mystery Game Puzzles
Integrity rank #14 of 65 · 76 models scored · top score 82.37 · GPT-6 Astra
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-6 Astra | 82.37 | unverifiedT2 | 2026-09-03 |
| 2 | GPT-6.1 Sol | 77.97 | unverifiedT2 | 2026-09-29 |
| 3 | Claude Opus 5.5 | 68.05 | unverifiedT2 | 2026-09-22 |
| 4 | Claude Sonnet 5.5 | 61.45 | unverifiedT2 | 2026-09-28 |
| 5 | Gemini 4 Argon | 60.34 | unverifiedT2 | 2026-09-30 |
| 6 | Claude Opus 5 | 54.84 | reproducedT1 | 2026-07-24 |
| 7 | Claude Fable 5.1 | 53.73 | unverifiedT2 | 2026-09-01 |
| 8 | GPT-5.6 Sol | 53.73 | unverifiedT2 | 2026-07-09 |
| 9 | GPT-6 Sol | 51.53 | unverifiedT2 | 2026-09-22 |
| 10 | GPT-5.5 | 51.53 | unverifiedT2 | 2026-04-23 |
| 11 | Claude Fable 5 | 47.12 | reproducedT1 | 2026-06-09 |
| 12 | Gemini 3.8 Flash | 41.62 | unverifiedT2 | 2026-09-02 |
| 13 | DeepSeek V4 Pro 0813 | 37.21 | reproducedT1 | 2026-08-13 |
| 14 | DeepSeek V4.1 Flash | 37.21 | unverifiedT2 | 2026-09-09 |
| 15 | Qwen 3.8 Max | 31.7 | reproducedT1 | 2026-08-02 |
| 16 | Gemini 3.7 Flash | 30.6 | reproducedT1 | 2026-08-13 |
| 17 | GPT-5.4 | 30.6 | unverifiedT2 | 2026-03-05 |
| 18 | Claude Opus 4.8 | 29.5 | reproducedT1 | 2026-05-28 |
| 19 | Claude Sonnet 5 | 28.4 | reproducedT1 | 2026-06-30 |
| 20 | GPT-5.6 Terra | 28.4 | unverifiedT2 | 2026-07-09 |
| 21 | Gemini 3.1 Pro | 27.3 | reproducedT1 | 2026-02-19 |
| 22 | Grok 4.6 | 27.3 | reproducedT1 | 2026-08-12 |
| 23 | DeepSeek V4 Flash 0731 | 27.3 | reproducedT1 | 2026-07-31 |
| 24 | GLM-5.3 | 26.2 | unverifiedT2 | 2026-08-14 |
| 25 | Gemini 3.5 Flash | 25.09 | reproducedT1 | 2026-05-19 |
| 26 | Qwen3.7-Max | 25.09 | unverifiedT2 | 2026-05-19 |
| 27 | Gemini 3.6 Flash | 22.89 | reproducedT1 | 2026-07-21 |
| 28 | Claude Haiku 5.5 | 22.89 | unverifiedT2 | 2026-10-07 |
| 29 | Grok 4.7 | 21.79 | unverifiedT2 | 2026-09-21 |
| 30 | o3 | 21.79 | unverifiedT2 | 2025-04-16 |
| 31 | Claude Opus 4.7 | 20.69 | reproducedT1 | 2026-04-16 |
| 32 | Gemini 3 Flash | 18.48 | reproducedT1 | 2025-12-17 |
| 33 | Kimi K3 | 18.48 | unverifiedT2 | 2026-07-16 |
| 34 | Muse Spark 1.3 | 17.38 | unverifiedT2 | 2026-09-02 |
| 35 | Claude Opus 4.6 | 17.38 | reproducedT1 | 2026-02-05 |
| 36 | GPT-5 | 15.18 | unverifiedT2 | 2025-08-07 |
| 37 | GPT-5.2 | 15.18 | unverifiedT2 | 2025-12-11 |
| 38 | Qwen 3.6 35B-A3B | 14.08 | unverifiedT2 | 2026-04-14 |
| 39 | Claude Opus 4.5 | 14.08 | reproducedT1 | 2025-11-24 |
| 40 | Claude Opus 4.1 | 12.98 | reproducedT1 | 2025-08-05 |
| 41 | GPT-5.6 Luna | 12.98 | unverifiedT2 | 2026-07-09 |
| 42 | Nemotron 3 Ultra | 11.87 | unverifiedT2 | 2026-06-04 |
| 43 | Qwen 3.5 Flash (hosted 35B-A3B) | 11.87 | unverifiedT2 | 2026-02-25 |
| 44 | GLM-5.2 | 10.77 | unverifiedT2 | 2026-06-16 |
| 45 | Gemini 3.5 Flash-Lite | 10.77 | reproducedT1 | 2026-07-21 |
| 46 | Qwen 3.6 Max (Preview) | 10.77 | unverifiedT2 | 2026-04-20 |
| 47 | GPT-5.1 | 10.77 | unverifiedT2 | 2025-11-13 |
| 48 | Kimi K2.6 | 9.67 | unverifiedT2 | 2026-04-20 |
| 49 | Qwen 3.6 Flash | 9.67 | unverifiedT2 | 2026-04-27 |
| 50 | Qwen3.5 397B-A17B | 9.67 | unverifiedT2 | 2026-02-13 |
| 51 | DeepSeek-V4-Pro | 8.57 | unverifiedT2 | 2026-04-24 |
| 52 | Mistral Large 4 | 8.57 | unverifiedT2 | 2026-10-06 |
| 53 | Qwen 3.5 Plus (hosted 397B-A17B) | 8.57 | unverifiedT2 | 2026-02-16 |
| 54 | Qwen3.7-Plus | 8.57 | unverifiedT2 | 2026-06-02 |
| 55 | Claude Sonnet 4.5 | 8.57 | reproducedT1 | 2025-09-29 |
| 56 | Claude Sonnet 4.6 | 7.47 | reproducedT1 | 2026-02-17 |
| 57 | Qwen3.7 Flash | 6.37 | unverifiedT2 | 2026-07-27 |
| 58 | Qwen 3.6 Plus | 3.06 | unverifiedT2 | 2026-03-31 |
| 59 | GPT-4 (Jun 2023) | 3.06 | unverifiedT2 | 2023-06-13 |
| 60 | GPT-4o mini | 3.06 | unverifiedT2 | 2024-07-18 |
| 61 | GPT-5.4 Mini | 1.96 | unverifiedT2 | 2026-03-17 |
| 62 | GPT-5 mini | 0.86 | unverifiedT2 | 2025-08-07 |
| 63 | gpt-oss-120b | 0 | unverifiedT2 | 2025-08-05 |
| 64 | GPT-6 Luna | 0 | unverifiedT2 | 2026-09-22 |
| 65 | Inkling-Small | 0 | reproducedT1 | 2026-07-15 |
| 66 | MiniMax-M3 | 0 | unverifiedT2 | 2026-06-01 |
| 67 | Qwen3-235B-A22B-Thinking (Jul 2025) | 0 | unverifiedT2 | 2025-07-25 |
| 68 | Qwen3-Max | 0 | unverifiedT2 | 2025-09-24 |
| 69 | Qwen3.6 27B | 0 | unverifiedT2 | 2026-04-22 |
| 70 | o3-mini | 0 | unverifiedT2 | 2025-01-31 |
| 71 | o4-mini | 0 | unverifiedT2 | 2025-04-16 |
| 72 | GLM-5.3-Flash | 0 | unverifiedT2 | 2026-08-20 |
| 73 | GPT-3.5 Turbo (Jan 2024) | 0 | unverifiedT2 | 2024-01-25 |
| 74 | GPT-4.1 mini | 0 | unverifiedT2 | 2025-04-14 |
| 75 | GPT-5 nano | 0 | unverifiedT2 | 2025-08-07 |
| 76 | GPT-5.4 Nano | 0 | unverifiedT2 | 2026-03-17 |
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.
The recorded metadata does not specify what this benchmark measures.
1 cited fact
This benchmark has scores from 76 tracked models, with a recorded score range spanning 82.37.
2 cited facts
Its Benchmark Integrity Index score is 96, with grade A and rank 14 out of 65 tracked benchmarks.
4 cited facts
It is not flagged as saturated. The top recorded score is 82.37, with 17.63 points of headroom. The computed top-score cluster contains 1 models.
4 cited facts
The harness classification is consistent harness. Test-set privacy is recorded as unknown. The recorded contamination-window ratio is 0.276.
3 cited facts
Mystery Game Puzzles: Mystery Game Puzzles as reported in Epoch AI's Capabilities Index CSV.
GPT-6 Astra leads Mystery Game Puzzles at 82.37. The full leaderboard above lists every recorded measurement, not just the headline number.
76 models have recorded scores on Mystery Game Puzzles, spanning a score spread of 82.37.
tensor.news grades Mystery Game Puzzles A for integrity (score 96/100), ranking #14 of 65 benchmarks we assess across discrimination, saturation, contamination resistance, harness comparability, and freshness.
No — Mystery Game Puzzles 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 Mystery Game Puzzles are reasonably apples-to-apples.
Every Mystery Game Puzzles measurement is source-backed and tagged reproduced, self-reported, or unverified; the underlying record cites epoch.ai/data/eci_benchmarks.csv.