Model benchmarks
Qwen3.8-Flash-Next-Q6Experts-g128-Q6Attn-g64-BF16-PLE-MLX-MTP-Q8 local LLM performance
As of September 2026, Qwen3.8-Flash-Next-Q6Experts-g128-Q6Attn-g64-BF16-PLE-MLX-MTP-Q8 runs at up to 64.0 tok/s for local inference (best of 3 community benchmark runs across 1 GPU).
Model size
4B
Peak speed
64.0 tok/s
Average speed
59.5 tok/s
Avg prefill
396.1 tok/s
Min memory
99.0 GB
Max context
262,144 tokens
Avg output / run
20,246 tokens
Avg runtime / run
6m 25s
Avg quality
81.3
Benchmark runs
3
GPUs tested
1
Quality by task
Average LLM-judged quality (0–100) with the run-to-run spread shown as a P5–P95 band, overall and for each benchmark task, across all 3 runs. The low and high columns show how much the judge’s score varies between runs, and need at least two runs to display.
| Task | P5 (low) | Avg | P95 (high) |
|---|---|---|---|
| Overall | 77.6 | 81.3 | 84.6 |
| Agent Workflow | 78.1 | 84.3 | 88.2 |
| Code Generation | 54.9 | 69.3 | 79.3 |
| Role Play & Narrative | 90.2 | 91.5 | 92.4 |
| Research & Analysis | 77.8 | 80.0 | 81.8 |
Performance by hardware and tool
Every hardware/tool/quantization combination Qwen3.8-Flash-Next-Q6Experts-g128-Q6Attn-g64-BF16-PLE-MLX-MTP-Q8 has been benchmarked on, ranked by peak token generation speed. Last updated September 2026.
| Hardware | Tool | Quant | Best tok/s | Avg tok/s | Memory | Context | Quality | Runs |
|---|---|---|---|---|---|---|---|---|
| Apple M4 Max | oMLX | — | 64.0 tok/s | 59.5 tok/s | 99.0 GB | 262,144 tokens | 81.3 | 3 |
Frequently asked questions
- Is Qwen3.8-Flash-Next-Q6Experts-g128-Q6Attn-g64-BF16-PLE-MLX-MTP-Q8 good for coding?
- In our benchmarks, Qwen3.8-Flash-Next-Q6Experts-g128-Q6Attn-g64-BF16-PLE-MLX-MTP-Q8 scores 69.3/100 for coding. It runs at about 59.5 tok/s, so if you want more speed, unsloth/Qwen3.8-27B-GGUF:IQ3_S is faster (~104.3 tok/s) and still scores well for coding (83.0/100).
- Is Qwen3.8-Flash-Next-Q6Experts-g128-Q6Attn-g64-BF16-PLE-MLX-MTP-Q8 good for agentic (tool-using) tasks?
- In our benchmarks, Qwen3.8-Flash-Next-Q6Experts-g128-Q6Attn-g64-BF16-PLE-MLX-MTP-Q8 scores 84.3/100 for agentic workflows. It runs at about 59.5 tok/s, so if you want more speed, granite-4.2-8b-Q4_K_L is faster (~71.6 tok/s) and still scores well for agentic workflows (92.9/100).
- How fast is Qwen3.8-Flash-Next-Q6Experts-g128-Q6Attn-g64-BF16-PLE-MLX-MTP-Q8 for local inference?
- Across 3 community benchmark runs, Qwen3.8-Flash-Next-Q6Experts-g128-Q6Attn-g64-BF16-PLE-MLX-MTP-Q8 reaches up to 64.0 tok/s and averages 59.5 tok/s, with the fastest results on Apple M4 Max.
- How much memory does Qwen3.8-Flash-Next-Q6Experts-g128-Q6Attn-g64-BF16-PLE-MLX-MTP-Q8 need?
- The leanest observed configuration used about 99.0 GB of memory.
- Which tools have been used to run Qwen3.8-Flash-Next-Q6Experts-g128-Q6Attn-g64-BF16-PLE-MLX-MTP-Q8?
- Benchmarks were submitted using oMLX. Results are community-contributed and updated as new runs arrive.