Benchmarks · composite
Vals RSI Index v1.1: Pre-training: Compression
Can a model do the research that builds the next model? Autonomous LLM R&D scored against frozen published, human, and frontier-model references. Published Pre-training: Compression score in percentage points.
21 models · 21 results · updated 2026-10-15 · vals.ai ↗
Leaderboard
Opus 5.5 max
More from this benchmark
6 published measuresScore vs any published measure6 measures: sub-scores, splits, counts
Breakdown
darker is better within each column · click a column to sort| Model | |||||
|---|---|---|---|---|---|
| Gemini 4 Argon max | 152.2M | 1.4M | 597k | 137.8M | 0 |
| GPT-6.1 Sol max | 209.6M | 2.3M | 1.9M | 196.8M | 11.7M |
| GPT-6 Sol max | 1273.7M | 1.2M | 602k | 1263.3M | 10.2M |
| Grok 4.7 max | 235.1M | 1.4M | 847k | 218.0M | 0 |
All results
21 rows| # | Model | Effort | Score |
|---|---|---|---|
| 1 | Opus 5.5 | max | 46.5% |
| 2 | Gemini 4 Argon | max | 44.7% |
| 3 | Fable 5.1 | max | 44.3% |
| 4 | GPT-6.1 Sol | max | 43.5% |
| 5 | GPT-6 Astra | max | 42.6% |
| 6 | Opus 5 | max | 42.6% |
| 7 | GPT-6 Sol | max | 42.3% |
| 8 | GPT-5.6 Sol | max | 41.0% |
| 9 | Fable 5 | max | 38.3% |
| 10 | Opus 4.8 | max | 38.3% |
| 11 | Grok 4.6 | max | 37.8% |
| 12 | Qwen 3.8 Max | max | 35.3% |
| 13 | GLM 5.3 | max | 33.4% |
| 14 | Kimi K3 | max | 33.3% |
| 15 | Opus 4.7 | max | 31.4% |
| 16 | GPT-5.5 | xhigh | 29.2% |
| 17 | Gemini 3.8 Flash | max | 29.0% |
| 18 | Grok 4.7 | max | 27.9% |
| 19 | GPT-5.4 | xhigh | 25.2% |
| 20 | Gemini 3.7 Flash | max | 25.0% |
| 21 | Muse Spark 1.3 | max | 23.1% |