Is unsloth/Qwen3.5-9B-MTP-GGUF good for coding?

In our benchmarks, unsloth/Qwen3.5-9B-MTP-GGUF scores 56.7/100 for coding. It runs at about 85.6 tok/s, so if you want more speed, Ornith-1.5-35B-A3B.APEX-I-NanoPlus is faster (~110.5 tok/s) and still scores well for coding (81.1/100).

Is unsloth/Qwen3.5-9B-MTP-GGUF good for agentic (tool-using) tasks?

In our benchmarks, unsloth/Qwen3.5-9B-MTP-GGUF scores 68.1/100 for agentic workflows. It runs at about 85.6 tok/s, so if you want more speed, ornith-1.5:35B is faster (~115.1 tok/s) and still scores well for agentic workflows (88.4/100).

How fast is unsloth/Qwen3.5-9B-MTP-GGUF for local inference?

Across 10 community benchmark runs, unsloth/Qwen3.5-9B-MTP-GGUF reaches up to 110.8 tok/s and averages 85.6 tok/s, with the fastest results on NVIDIA GeForce RTX 4070.

How much memory does unsloth/Qwen3.5-9B-MTP-GGUF need?

The leanest observed configuration used about 10.1 GB of memory (quantizations tested: Q6_K, UD-Q4_K_XL, UD-Q5_K_XL).

Which tools have been used to run unsloth/Qwen3.5-9B-MTP-GGUF?

Benchmarks were submitted using Unsloth Studio, llama.cpp. Results are community-contributed and updated as new runs arrive.

Model benchmarks

unsloth/Qwen3.5-9B-MTP-GGUF local LLM performance

As of October 2026, unsloth/Qwen3.5-9B-MTP-GGUF runs at up to 110.8 tok/s for local inference (best of 10 community benchmark runs across 1 GPU).

Unsloth Studiollama.cppQ6_KUD-Q4_K_XLUD-Q5_K_XL

ShareReddit X

Model size

Peak speed

110.8 tok/s

Average speed

85.6 tok/s

Avg PP

1333.0 tok/s

Min memory

10.1 GB

Max context

262,144 tokens

Avg output / run

11,850 tokens

Avg runtime / run

2m 51s

Avg quality

68.6

Benchmark runs

GPUs tested

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 10 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	63.9	68.6	72.5
Agent Workflow	52.6	68.1	76.9
Code Generation	43.2	56.7	63.1
Role Play & Narrative	71.9	78.1	85.1
Research & Analysis	62.9	71.6	79.3

Performance by hardware and tool

Every hardware/tool/quantization combination unsloth/Qwen3.5-9B-MTP-GGUF has been benchmarked on, ranked by peak token generation speed. Last updated October 2026.

Hardware	Tool	Quant	Best tok/s	Avg tok/s	Memory	Context	Quality	Runs
NVIDIA GeForce RTX 4070	Unsloth Studio	UD-Q4_K_XL	110.8 tok/s	84.7 tok/s	n/a	262,144 tokens	68.3	4
NVIDIA GeForce RTX 4070	llama.cpp	—	102.9 tok/s	99.3 tok/s	n/a	262,144 tokens	69.0	2
NVIDIA GeForce RTX 4070	Unsloth Studio	UD-Q5_K_XL	102.0 tok/s	87.6 tok/s	10.1 GB	65,536 tokens	68.4	3
NVIDIA GeForce RTX 4070	Unsloth Studio	Q6_K	56.0 tok/s	56.0 tok/s	10.3 GB	65,536 tokens	69.7	1

Benchmark runs

All 10 unsloth/Qwen3.5-9B-MTP-GGUF runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.

Open run page →

Frequently asked questions

Is unsloth/Qwen3.5-9B-MTP-GGUF good for coding?: In our benchmarks, unsloth/Qwen3.5-9B-MTP-GGUF scores 56.7/100 for coding. It runs at about 85.6 tok/s, so if you want more speed, Ornith-1.5-35B-A3B.APEX-I-NanoPlus is faster (~110.5 tok/s) and still scores well for coding (81.1/100).
Is unsloth/Qwen3.5-9B-MTP-GGUF good for agentic (tool-using) tasks?: In our benchmarks, unsloth/Qwen3.5-9B-MTP-GGUF scores 68.1/100 for agentic workflows. It runs at about 85.6 tok/s, so if you want more speed, ornith-1.5:35B is faster (~115.1 tok/s) and still scores well for agentic workflows (88.4/100).
How fast is unsloth/Qwen3.5-9B-MTP-GGUF for local inference?: Across 10 community benchmark runs, unsloth/Qwen3.5-9B-MTP-GGUF reaches up to 110.8 tok/s and averages 85.6 tok/s, with the fastest results on NVIDIA GeForce RTX 4070.
How much memory does unsloth/Qwen3.5-9B-MTP-GGUF need?: The leanest observed configuration used about 10.1 GB of memory (quantizations tested: Q6_K, UD-Q4_K_XL, UD-Q5_K_XL).
Which tools have been used to run unsloth/Qwen3.5-9B-MTP-GGUF?: Benchmarks were submitted using Unsloth Studio, llama.cpp. Results are community-contributed and updated as new runs arrive.