← Back to community benchmarks
gemma-4-26B-A4B-it-uncensored-heretic-ara8bit-int8-affine
Performance
32k
tokens
1,513
PP tok/s
59.8
TG tok/s
21659
TTFT (ms)
28.9
Peak mem (GB)
Hardware
Chip
M3 Ultra (60c)
Memory
256 GB
GPU Cores
60
Software
oMLX
v0.3.12
macOS
macOS 26.4.1
Context
32,768
Performance by Context Length
| Context | PP tok/s | TG tok/s | Peak Mem | |
|---|---|---|---|---|
| 1k | 1,315 | 80.9 | 27.2 GB | view |
| 4k | 1,679 | 78.6 | 27.6 GB | view |
| 8k | 1,730 | 74.7 | 27.7 GB | view |
| 16k | 1,667 | 69.2 | 28.1 GB | view |
| 32k | 1,513 | 59.8 | 28.9 GB | current |
| 64k | 1,238 | 46.0 | 30.6 GB | view |
| 128k | 889.5 | 30.8 | 34.2 GB | view |
Batching Results
| Batch Size | TG tok/s | Speedup |
|---|---|---|
| 1× | 80.9 | 1.00× |
| 2× | 117.2 | 1.45× |
| 4× | 162.0 | 2.00× |
| 8× | 223.5 | 2.76× |