The Latent Score: 2026-09-04
Release tli-2026-v4.1-2026-09-04b. Methodology tli-2026-v4.1.
As of 2026-09-04, Fable 5.1 leads The Latent Score on the point estimate at 128.3 points among 19 ranked models. Rankings describe published capability evidence, not a statistically proven winner or the best model for every workload.
This page preserves a specific publication. Its data does not change when a new release is published.
Published rankings
| Rank | Model and evidence | Score (points) | Stability range (points) | Contributing results |
|---|---|---|---|---|
| 1 | Fable 5.1 | 128.3 | 122.7–134.2 | 50 |
| 2 | Opus 5 | 121.5 | 114.8–126.4 | 76 |
| 3 | GPT-6 Astra | 121.4 | 110.6–146.7 | 32 |
| 4 | Fable 5 | 118.2 | 112.9–122.9 | 84 |
| 5 | Muse Spark 1.3 | 112.4 | 105.6–119.6 | 18 |
| 6 | GPT-5.6 Sol | 107.8 | 103.4–114.2 | 87 |
| 7 | Kimi K3 | 105.2 | 98.7–109.7 | 75 |
| 8 | GLM-5.3 | 103.7 | 95.2–108.9 | 48 |
| 9 | Muse Spark 1.2 | 102.6 | 93.4–107.8 | 44 |
| 10 | Gemini 3.8 Flash | 102.3 | 93.7–110.5 | 42 |
| 11 | Grok 4.6 | 101.1 | 96–107.8 | 49 |
| 12 | Qwen3.8-Max | 97.3 | 89.7–101.5 | 61 |
| 13 | Gemini 3.7 Flash | 94.1 | 84.9–104 | 52 |
| 14 | GPT-5.6 Terra | 93.9 | 89.7–101.3 | 64 |
| 15 | Sonnet 5 | 88.6 | 82.3–94 | 62 |
| 16 | DeepSeek V4 Pro | 87.8 | 81.2–97.1 | 49 |
| 17 | GPT-5.6 Luna | 86.2 | 82.1–93.5 | 61 |
| 18 | DeepSeek V4 Flash | 72.5 | 67–81.9 | 49 |
| 19 | Qwen3.8-27B | 70.7 | 61.8–80.8 | 44 |
Methodology and limitations in this release
The scale is anchored to mean 100, SD 15 across 18 frozen reference models. It is not human IQ or a percentage. 88 benchmarks contribute across 7 domains.
This older release did not preserve a written methodology in its public snapshot. Its recorded scores and scale are archived here; current methodology documentation must not be assumed to describe this historical calculation.
Reading this release
Compare model scores alongside their stability ranges and domain coverage. A higher point estimate does not establish superiority for every task. The linked model records disclose contributing and excluded results and their published configurations.
This release uses methodology version tli-2026-v4.1. Comparisons with another release should check both its methodology version and evidence date; a new data release does not necessarily mean the scoring method changed.
Citation and downloads
The Latent. “The Latent Score.” Release tli-2026-v4.1-2026-09-04b, dataset dated 2026-09-04. Permanent release page.
Download public scores with definitions (JSON) · Data usage terms