Model benchmarks
Huihui-Qwen3-Coder-30B-A3B-Instruct-abliterated.i1-Q4_K_M local LLM performance
As of November 2026, Huihui-Qwen3-Coder-30B-A3B-Instruct-abliterated.i1-Q4_K_M runs at up to 110.1 tok/s for local inference (best of 4 community benchmark runs across 0 GPUs).
Model size
30B
Peak speed
110.1 tok/s
Average speed
109.9 tok/s
Avg PP
2596.8 tok/s
Min memory
47.8 GB
Max context
262,144 tokens
Avg output / run
6,902 tokens
Avg runtime / run
1m 9s
Avg quality
59.7
Benchmark runs
4
GPUs tested
0
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 4 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 | 57.2 | 59.7 | 64.5 |
| Agent Workflow | 36.3 | 54.3 | 67.3 |
| Code Generation | 57.5 | 65.8 | 71.4 |
| Role Play & Narrative | 52.8 | 65.3 | 72.3 |
| Research & Analysis | 50.5 | 53.6 | 56.2 |
Performance by hardware and tool
Every hardware/tool/quantization combination Huihui-Qwen3-Coder-30B-A3B-Instruct-abliterated.i1-Q4_K_M has been benchmarked on, ranked by peak token generation speed. Last updated November 2026.
| Hardware | Tool | Quant | Best tok/s | Avg tok/s | Memory | Context | Quality | Runs |
|---|---|---|---|---|---|---|---|---|
| CPU only | llama.cpp | Q4_K_M | 110.1 tok/s | 109.9 tok/s | 47.8 GB | 262,144 tokens | 59.7 | 4 |
Benchmark runs
All 4 Huihui-Qwen3-Coder-30B-A3B-Instruct-abliterated.i1-Q4_K_M runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.
Frequently asked questions
- Is Huihui-Qwen3-Coder-30B-A3B-Instruct-abliterated.i1-Q4_K_M good for coding?
- In our benchmarks, Huihui-Qwen3-Coder-30B-A3B-Instruct-abliterated.i1-Q4_K_M scores 65.8/100 for coding. It runs at about 109.9 tok/s, so if you want more speed, ornith-1.5-35b-apex is faster (~149.2 tok/s) and still scores well for coding (83.5/100).
- Is Huihui-Qwen3-Coder-30B-A3B-Instruct-abliterated.i1-Q4_K_M good for agentic (tool-using) tasks?
- In our benchmarks, Huihui-Qwen3-Coder-30B-A3B-Instruct-abliterated.i1-Q4_K_M scores 54.3/100 for agentic workflows. It runs at about 109.9 tok/s, so if you want more speed, Qwen3.8-27B-IQ5R-V4-DFlash2-TP2-i8KV-r21 is faster (~133.3 tok/s) and still scores well for agentic workflows (91.8/100).
- How fast is Huihui-Qwen3-Coder-30B-A3B-Instruct-abliterated.i1-Q4_K_M for local inference?
- Across 4 community benchmark runs, Huihui-Qwen3-Coder-30B-A3B-Instruct-abliterated.i1-Q4_K_M reaches up to 110.1 tok/s and averages 109.9 tok/s.
- How much memory does Huihui-Qwen3-Coder-30B-A3B-Instruct-abliterated.i1-Q4_K_M need?
- The leanest observed configuration used about 47.8 GB of memory (quantizations tested: Q4_K_M).
- Which tools have been used to run Huihui-Qwen3-Coder-30B-A3B-Instruct-abliterated.i1-Q4_K_M?
- Benchmarks were submitted using llama.cpp. Results are community-contributed and updated as new runs arrive.