Vals RSI — Compression
Agents · 2026-09-21
Reference-normalized score for lossless compression of visible enwik8 within 24 hours, one H100 and eight CPU cores offline. A clean sandbox verifies exact reconstruction; artifact size includes decompressor dependencies. Baseline 2.92 BPB, reference 1.2435, theoretical best 0.6.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5.5 | 46.5 |
| Claude Fable 5.1 | 44.3 |
| GPT-6 Astra | 42.6 |
| Claude Opus 5 | 42.6 |
| GPT-5.6 Sol | 41.0 |
| Claude Fable 5 | 38.3 |
| Opus 4.8 | 38.3 |
| Grok 4.6 | 37.8 |
| GLM-5.3 | 33.4 |
| Kimi K3 | 33.3 |
| Opus 4.7 | 31.4 |
| GPT-5.5 | 29.2 |
| Gemini 3.8 Flash | 29.0 |
| GPT-5.2 | 28.7 |
| GPT-5.4 | 25.2 |
| Muse Spark 1.3 | 23.1 |
| Kimi K2.5 | 0.0 |