Add live API demo results (January 2026)
Browse files
README.md
CHANGED
|
@@ -59,10 +59,27 @@ Measures descriptive language richness.
|
|
| 59 |
|
| 60 |
---
|
| 61 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 62 |
## The Unlock
|
| 63 |
|
| 64 |
We're not claiming to beat Claude on language quality.
|
| 65 |
-
We're claiming to **match their quality** while running **
|
| 66 |
|
| 67 |
That enables:
|
| 68 |
- Live broadcasting with real-time audio description
|
|
|
|
| 59 |
|
| 60 |
---
|
| 61 |
|
| 62 |
+
## Live API Demo Results (January 2026)
|
| 63 |
+
|
| 64 |
+
We built a live demo that races Visual Narrator against frontier models using **real API calls**—no simulation, no cherry-picking.
|
| 65 |
+
|
| 66 |
+
| Model | Live Latency | vs Visual Narrator |
|
| 67 |
+
|-------|-------------|-------------------|
|
| 68 |
+
| **Visual Narrator** | **429ms** | — |
|
| 69 |
+
| Claude Sonnet 4 | 4,559ms | 10.6x slower |
|
| 70 |
+
| Gemini 2.0 Flash | 8,048ms | 18.8x slower |
|
| 71 |
+
| GPT-4o | 11,873ms | 27.7x slower |
|
| 72 |
+
|
| 73 |
+
**Try it yourself:** [Live Comparison Demo](https://huggingface.co/spaces/Ytgetahun/visual-narrator-comparison)
|
| 74 |
+
|
| 75 |
+
*Results from parallel API calls at the same millisecond. WebSocket endpoint available for verification.*
|
| 76 |
+
|
| 77 |
+
---
|
| 78 |
+
|
| 79 |
## The Unlock
|
| 80 |
|
| 81 |
We're not claiming to beat Claude on language quality.
|
| 82 |
+
We're claiming to **match their quality** while running **10x+ faster in real-world API conditions**.
|
| 83 |
|
| 84 |
That enables:
|
| 85 |
- Live broadcasting with real-time audio description
|