Model benchmarks
minicpm-v-4.7-35b-a3b-q6_k local LLM performance
As of November 2026, minicpm-v-4.7-35b-a3b-q6_k runs at up to 31.1 tok/s for local inference (best of 2 community benchmark runs across 1 GPU).
Model size
35B
Peak speed
31.1 tok/s
Average speed
22.0 tok/s
Avg PP
312.3 tok/s
Min memory
16.8 GB
Max context
65,536 tokens
Avg output / run
8,027 tokens
Avg runtime / run
7m 47s
Avg quality
77.4
Benchmark runs
2
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 2 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 | 75.9 | 77.4 | 79.0 |
| Agent Workflow | 86.3 | 87.8 | 89.2 |
| Code Generation | 52.6 | 60.6 | 68.5 |
| Role Play & Narrative | 89.1 | 89.3 | 89.4 |
| Research & Analysis | 69.3 | 72.3 | 75.2 |
Performance by hardware and tool
Every hardware/tool/quantization combination minicpm-v-4.7-35b-a3b-q6_k 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 |
|---|---|---|---|---|---|---|---|---|
| AMD Radeon RX 7800 XT | llama.cpp | Q6_K | 31.1 tok/s | 22.0 tok/s | 16.8 GB | 65,536 tokens | 77.5 | 2 |
Benchmark runs
All 2 minicpm-v-4.7-35b-a3b-q6_k runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.
Frequently asked questions
- Is minicpm-v-4.7-35b-a3b-q6_k good for coding?
- In our benchmarks, minicpm-v-4.7-35b-a3b-q6_k scores 60.6/100 for coding. It runs at about 22.0 tok/s, so if you want more speed, Huihui-Qwen3.8-27B-abliterated-Swift-1.5-GSQ-RCO-IQ3_S-mtp is faster (~43.5 tok/s) and still scores well for coding (86.6/100).
- Is minicpm-v-4.7-35b-a3b-q6_k good for agentic (tool-using) tasks?
- In our benchmarks, minicpm-v-4.7-35b-a3b-q6_k scores 87.8/100 for agentic workflows. It runs at about 22.0 tok/s, so if you want more speed, muse-glimmer:latest is faster (~33.3 tok/s) and still scores well for agentic workflows (91.6/100).
- How fast is minicpm-v-4.7-35b-a3b-q6_k for local inference?
- Across 2 community benchmark runs, minicpm-v-4.7-35b-a3b-q6_k reaches up to 31.1 tok/s and averages 22.0 tok/s, with the fastest results on AMD Radeon RX 7800 XT.
- How much memory does minicpm-v-4.7-35b-a3b-q6_k need?
- The leanest observed configuration used about 16.8 GB of memory (quantizations tested: Q6_K).
- Which tools have been used to run minicpm-v-4.7-35b-a3b-q6_k?
- Benchmarks were submitted using llama.cpp. Results are community-contributed and updated as new runs arrive.