Capability breakdown
Intelligence over time
Provider & region options
Fastest loop: Overshoot · US East (Virginia) · 141 ms end to end.
| Overshoot | US East (Virginia) | 270 tok/s | 52 ms | 141 ms | $0.27 | <200 ms |
| Overshoot | US West (Oregon) | 270 tok/s | 54 ms | 143 ms | $0.27 | <200 ms |
| Overshoot | EU West (Ireland) | 270 tok/s | 63 ms | 152 ms | $0.27 | <200 ms |
| Overshoot | EU Central (Frankfurt) | 270 tok/s | 67 ms | 155 ms | $0.27 | <200 ms |
| Overshoot | Asia Pacific (Tokyo) | 270 tok/s | 80 ms | 169 ms | $0.27 | <200 ms |
| Overshoot | Asia Pacific (Mumbai) | 270 tok/s | 100 ms | 189 ms | $0.27 | <200 ms |
| Overshoot | Middle East (Dubai) | 270 tok/s | 103 ms | 192 ms | $0.27 | <200 ms |
| Overshoot | South America (São Paulo) | 270 tok/s | 108 ms | 197 ms | $0.27 | <200 ms |
| Together | US East (Virginia) | 173 tok/s | 91 ms | 230 ms | $0.27 | – |
| Together | US West (Oregon) | 173 tok/s | 97 ms | 236 ms | $0.27 | – |
| Together | EU West (Ireland) | 173 tok/s | 103 ms | 242 ms | $0.27 | – |
| Novita | US East (Virginia) | 150 tok/s | 98 ms | 259 ms | $0.21 | – |
| vLLM (self-host) | US East (Virginia) | 137 tok/s | 106 ms | 281 ms | $0.17 | – |
| Novita | Asia Pacific (Tokyo) | 150 tok/s | 127 ms | 287 ms | $0.21 | – |
| vLLM (self-host) | EU West (Ireland) | 137 tok/s | 117 ms | 291 ms | $0.17 | – |
Sample outputs
Illustrative samples shared across models, not verbatim output from Gemma 4 31B.
Scene description
“Describe what's happening in this camera frame.”
A forklift is reversing toward loading bay 3 while a worker in a hi-vis vest signals from the left. Two pallets remain unwrapped near the dock edge.
OCR + extraction
“Read the shipping label and return structured JSON.”
{ "tracking": "1Z A24 9W7 03 8421 990 4", "carrier": "UPS", "weight_kg": 12.4, "dest": "Portland, OR 97204" }
Grounding
“Point to every person not wearing a hard hat.”
2 detections, [x:0.41, y:0.62, w:0.08, h:0.19] conf 0.94 · [x:0.77, y:0.55, w:0.07, h:0.21] conf 0.88.
Related models
Where it ranks
- #8Gemini 3 Flash75.3
- #9Claude Haiku 4.574.2
- #10Holo3 35B A3B74.0
- #11Gemma 4 31B73.0
- #12Gemma 4 26B A4B69.3
- #13GPT-5.4 nano63.6