Pemantau degradasi model v0.1
Data mentah hasil pengukuran satu model di berbagai relay: pada slot waktu yang sama, setiap relay menjawab set probe identitas yang sama (batas pengetahuan, “versi terbaru yang Anda ketahui”, dan sejenisnya), lalu kami mencatat kesesuaian dengan jawaban acuan model tersebut, token input, token penalaran, dan sidik jari model. Angka dipublikasikan apa adanya, tanpa vonis.
| 100% (90 jawaban) | 109–24,260 | 1,014 | DeepSeek V4.1 Flash | — | 98.2% | 1.31 s | ¥2.59 | ||
| 100% (90 jawaban) | 109–24,260 | 1,014 | DeepSeek V4 Pro | — | 99.0% | 1.35 s | ¥11.4 | ||
| 100% (90 jawaban) | 109–24,260 | 1,014 | GLM 5.3 | — | 98.5% | 0.96 s | ¥8.56 | ||
| 100% (90 jawaban) | 109–24,260 | 1,014 | Qwen 3.8 Max | — | 98.0% | 1.06 s | ¥20.4 | ||
| 98% (90 jawaban) | 109–24,260 | 0 | Claude Haiku 5.5 | — | — | — | — | ||
| 98% (90 jawaban) | 100–161,349 | 38 | GPT 6.1 Sol | — | 98.7% | 0.87 s | ¥0.44 | ||
| 98% (90 jawaban) | — | 0 | Grok 4.7 | — | 98.3% | 5.11 s | ¥3.10 | ||
| 97% (90 jawaban) | 52–24,260 | 996 | Claude Fable 5.1 | — | 98.2% | 1.34 s | ¥4.38 | ||
| 97% (90 jawaban) | 52–24,260 | 996 | Claude Fable 5.1 | — | — | — | — | ||
| 97% (90 jawaban) | 52–24,260 | 996 | Claude Fable 5.1 | — | — | — | — | ||
| 97% (90 jawaban) | 52–24,260 | 996 | Claude Fable 5.1 | — | — | — | — | ||
| 97% (90 jawaban) | 52–24,260 | 559 | Claude Opus 5.5 | — | 97.3% | 1.57 s | ¥1.75 | ||
| 97% (90 jawaban) | 52–24,260 | 559 | Claude Opus 5.5 | — | — | — | — | ||
| 97% (90 jawaban) | 52–24,260 | 559 | Claude Opus 5.5 | — | — | — | — | ||
| 97% (90 jawaban) | 109–45,108 | 0 | Claude Sonnet 5 | — | 97.7% | 1.77 s | ¥7.30 | ||
| 97% (90 jawaban) | 109–45,108 | 0 | Claude Sonnet 5 | — | 97.7% | 1.58 s | ¥9.20 | ||
| 97% (90 jawaban) | 109–45,108 | 0 | Claude Sonnet 5 | — | 97.2% | 1.29 s | ¥0.88 | ||
| 97% (90 jawaban) | 109–45,108 | 0 | Claude Sonnet 5 | — | — | — | — | ||
| 97% (90 jawaban) | 109–24,260 | 0 | Claude Haiku 5.5 | — | 98.0% | 0.94 s | — | ||
| 97% (90 jawaban) | 109–24,260 | 0 | Claude Haiku 5.5 | — | — | — | — | ||
| 97% (90 jawaban) | 100–151,769 | 39 | GPT 6 Astra | — | 98.5% | 1.33 s | ¥2.19 | ||
| 97% (90 jawaban) | 100–151,769 | 39 | GPT 6 Astra | — | — | — | — | ||
| 97% (90 jawaban) | 100–151,769 | 39 | GPT 6 Astra | — | — | — | — | ||
| 97% (90 jawaban) | 100–151,769 | 39 | GPT 6 Astra | — | — | — | — | ||
| 97% (90 jawaban) | 100–151,769 | 39 | GPT 6 Astra | — | — | — | — | ||
| 97% (90 jawaban) | 100–161,349 | 38 | GPT 6.1 Sol | — | — | — | — | ||
| 97% (90 jawaban) | 100–161,349 | 38 | GPT 6.1 Sol | — | — | — | — | ||
| 97% (90 jawaban) | 100–161,349 | 38 | GPT 6.1 Sol | — | — | — | — | ||
| 97% (90 jawaban) | — | 0 | Grok 4.7 | — | — | — | — | ||
| 97% (90 jawaban) | — | 0 | Kimi K3 | — | 98.0% | 1.16 s | ¥28.6 | ||
| 85% (20 jawaban) | 101–22,953 | 23 | GPT 6 Astra | — | 100.0% | 9.76 s | — | ||
| 61% (18 jawaban) | 116–402 | 217 | Claude Opus 5.5 | — | 93.3% | 9.05 s | — | ||
| 44% (18 jawaban) | 116–139 | 243 | Claude Haiku 5.5 | — | 96.7% | 7.00 s | — | ||
| 39% (18 jawaban) | 114–411 | 0 | Claude Haiku 5.5 | — | 100.0% | 23.19 s | — | ||
| 69% (13 jawaban) | 3,685–21,590 | 0 | Gemini 3.8 Flash | — | — | — | — | ||
| 54% (28 jawaban) | 121–140 | 0 | Grok 4.7 | — | — | — | — | ||
| 28% (18 jawaban) | 239–748 | 0 | Claude Haiku 5.5 | — | 77.4% | 19.67 s | — | ||
| 100% (54 jawaban) | 456–694 | 69 | Claude Opus 5.5 | — | — | — | — | ||
| 100% (54 jawaban) | 472–730 | 178 | Claude Haiku 5.5 | — | — | — | — | ||
| 100% (8 jawaban) | 22,945–22,956 | 38 | GPT 6 Astra | — | 100.0% | 4.04 s | ¥3.30 | ||
| 96% (28 jawaban) | 5,264–5,279 | 29 | GPT 6 Astra | — | — | — | — | ||
| 89% (28 jawaban) | 4,203–4,218 | 27 | GPT 6 Astra | — | 100.0% | 8.02 s | — | ||
| 89% (28 jawaban) | 486–4,212 | 35 | GPT 6 Astra | — | — | — | — | ||
| 89% (18 jawaban) | 129–155 | 0 | Claude Fable 5.1 | — | — | — | — | ||
| 88% (8 jawaban) | 22,945–22,956 | 42 | GPT 6.1 Sol | — | — | — | — | ||
| 86% (28 jawaban) | 93–4,218 | 18 | GPT 6 Astra | — | — | — | — | ||
| 83% (18 jawaban) | 127–153 | 0 | Claude Sonnet 5 | — | 96.7% | 6.55 s | — | ||
| 83% (18 jawaban) | 116–139 | 0 | Claude Haiku 5.5 | — | — | — | — | ||
| 82% (28 jawaban) | 93–108 | 36 | GPT 6 Astra | — | 100.0% | 2.88 s | — | ||
| 82% (28 jawaban) | 812–828 | 0 | GPT 6.1 Sol | — | 100.0% | 3.49 s | ¥15.4 | ||
| 82% (28 jawaban) | 100–116 | 38 | GPT 6.1 Sol | — | — | — | — | ||
| 79% (28 jawaban) | 93–108 | 33 | GPT 6 Astra | — | 100.0% | 9.99 s | ¥2.04 | ||
| 79% (28 jawaban) | 93–108 | 34 | GPT 6.1 Sol | — | — | — | — | ||
| 72% (18 jawaban) | 129–155 | 537 | Claude Haiku 5.5 | — | 77.8% | 14.16 s | — | ||
| 68% (28 jawaban) | 4,207–4,480 | 16 | GPT 6.1 Sol | — | 96.7% | 5.20 s | — | ||
| 68% (28 jawaban) | 101–4,484 | 0 | GPT 6 Astra | — | 93.3% | 8.16 s | ¥11.9 | ||
| 61% (28 jawaban) | 4,211–18,563 | 0 | GPT 5.6 Terra | — | — | — | — | ||
| 56% (18 jawaban) | 116–139 | 0 | Claude Opus 5.5 | — | 83.3% | 12.04 s | ¥8.80 | ||
| 56% (18 jawaban) | 114–137 | 0 | Claude Haiku 5.5 | — | 77.8% | 9.30 s | ¥2.04 | ||
| 50% (18 jawaban) | 2,025–10,740 | 0 | Claude Haiku 5.5 | — | 79.3% | 21.85 s | — | ||
| 50% (18 jawaban) | 127–153 | 458 | Claude Sonnet 5 | — | 61.1% | 12.08 s | — | ||
| 44% (18 jawaban) | 116–139 | 170 | Claude Haiku 5.5 | — | 100.0% | 2.17 s | — | ||
| 44% (18 jawaban) | 258–2,860 | 0 | Claude Haiku 5.5 | — | 88.9% | 24.58 s | ¥2.04 | ||
| 44% (18 jawaban) | 153–168 | 0 | Claude Sonnet 5 | — | 100.0% | 3.97 s | ¥4.40 | ||
| 41% (17 jawaban) | 116–139 | 0 | Claude Haiku 5.5 | — | — | — | — | ||
| 40% (5 jawaban) | 137–144 | 0 | Claude Opus 5.5 | — | 60.0% | 2.38 s | ¥4.77 | ||
| 39% (18 jawaban) | 116–139 | 208 | Claude Haiku 5.5 | — | 100.0% | 2.03 s | — | ||
| 33% (18 jawaban) | 121–239 | 0 | Claude Haiku 5.5 | — | 100.0% | 2.88 s | ¥4.40 | ||
| 28% (18 jawaban) | 148–463 | 0 | Claude Haiku 5.5 | — | 88.9% | 15.94 s | ¥2.04 | ||
| Belum diukur | — | — | — | — | 100.0% | 23.19 s | — | ||
| Belum diukur | — | — | — | — | 100.0% | 23.19 s | — | ||
| Belum diukur | — | — | — | — | 100.0% | 8.85 s | — | ||
| Belum diukur | — | — | — | — | 98.6% | 1.40 s | ¥3.29 | ||
| Belum diukur | — | — | — | — | 98.2% | 0.91 s | ¥4.46 | ||
| Belum diukur | — | — | — | — | 73.3% | 1.83 s | — | ||
| Belum diukur | — | — | — | — | 100.0% | 2.27 s | — | ||
| Belum diukur | — | — | — | — | 77.4% | 3.02 s | ¥4.40 | ||
| Belum diukur | — | — | — | — | 100.0% | 3.52 s | — | ||
| Belum diukur | — | — | — | — | 100.0% | 6.04 s | — | ||
| Belum diukur | — | — | — | — | 100.0% | 2.45 s | — | ||
| Belum diukur | — | — | — | — | 100.0% | 4.62 s | ¥2.38 | ||
| Belum diukur | — | — | — | — | 100.0% | 7.18 s | ¥0.66 | ||
| Belum diukur | — | — | — | — | 90.0% | 6.70 s | — | ||
| Belum diukur | — | — | — | — | 86.7% | 4.67 s | — | ||
| Belum diukur | — | — | — | — | 0.0% | — | — | ||
| Belum diukur | — | — | — | — | 100.0% | 2.51 s | — | ||
| Belum diukur | — | — | — | — | 100.0% | 3.63 s | — | ||
| Belum diukur | — | — | — | — | 100.0% | 3.95 s | — | ||
| Belum diukur | — | — | — | — | 100.0% | 4.17 s | ¥1.30 | ||
| Belum diukur | — | — | — | — | 94.4% | 6.48 s | — | ||
| Belum diukur | — | — | — | — | 89.7% | 6.03 s | — | ||
| Belum diukur | — | — | — | — | 96.7% | 9.57 s | — | ||
| Belum diukur | — | — | — | — | 88.9% | 6.48 s | ¥2.04 | ||
| Belum diukur | — | — | — | — | 100.0% | 3.31 s | — | ||
| Belum diukur | — | — | — | — | 100.0% | 4.88 s | ¥1.27 | ||
| Belum diukur | — | — | — | — | 100.0% | 12.91 s | — | ||
| Belum diukur | — | — | — | — | 96.7% | 10.99 s | — | ||
| Belum diukur | — | — | — | — | 0.0% | — | — | ||
| Belum diukur | — | — | — | — | 96.7% | 4.34 s | — | ||
| Belum diukur | — | — | — | — | 88.9% | 5.53 s | ¥2.04 | ||
| Belum diukur | — | — | — | — | 90.0% | 7.27 s | — |
Pertanyaan umum
Bagaimana cara membaca angka-angka ini?
Probe menanyakan hal seperti “kapan batas pengetahuan Anda” atau “iPhone terbaru mana yang Anda ketahui”: model yang berbeda menjawab berbeda, sedangkan satu model menjawab secara konsisten. Kesesuaian yang jauh lebih rendah dibanding relay lain, jumlah token input yang tidak cocok dengan relay lain, atau sidik jari yang tidak mirip model tersebut patut diperiksa lebih lanjut. Angka tunggal bisa berfluktuasi; makin banyak jawaban, makin andal angkanya.
Mengapa melihat jumlah token input?
Input yang sama yang dikirim ke model yang sama seharusnya selalu menghasilkan jumlah token yang persis sama. Jumlah yang berbeda dari relay lain biasanya berarti modelnya berbeda, atau ada konten tambahan yang disisipkan ke permintaan.
Mengapa belum ada data?
Pemantau degradasi model masih dalam proses integrasi. Hasil pengukuran pertama akan muncul di sini. Ketersediaan, latensi token pertama, dan biaya sesi di tabel adalah hasil pengukuran langsung.
Apa bedanya dengan peringkat utama?
Peringkat utama mencakup ketersediaan, kecepatan, dan harga. Pemantau degradasi model menunjukkan apakah setiap relay benar-benar menyajikan model yang asli. Untuk saat ini, pemantau ini tidak memengaruhi peringkat.