Model benchmarks

deepseek-r1-0528-qwen3-8b local LLM performance

As of September 2026, deepseek-r1-0528-qwen3-8b runs at up to 42.8 tok/s for local inference (best of 2 community benchmark runs across 1 GPU).

LM Studio
ShareRedditX

Model size

8B

Peak speed

42.8 tok/s

Average speed

37.3 tok/s

Avg PP

552.9 tok/s

Min memory

4.7 GB

Max context

65.536 tokens

Avg output / run

27.500 tokens

Avg runtime / run

13m 8s

Avg quality

30.6

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.

TaskP5 (low)AvgP95 (high)
Overall28.930.632.2
Agent Workflow29.445.160.9
Code Generation3.86.59.2
Role Play & Narrative45.257.770.3
Research & Analysis12.312.913.5

Performance by hardware and tool

Every hardware/tool/quantization combination deepseek-r1-0528-qwen3-8b has been benchmarked on, ranked by peak token generation speed. Last updated September 2026.

HardwareToolQuantBest tok/sAvg tok/sMemoryContextQualityRuns
Apple M4 MaxLM Studio42.8 tok/s42.8 tok/s4.7 GB65.536 tokens32.41
CPU onlyLM Studio31.9 tok/s31.9 tok/s4.7 GB8.192 tokens28.81

Benchmark runs

All 2 deepseek-r1-0528-qwen3-8b runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.

Frequently asked questions

Is deepseek/deepseek-r1-0528-qwen3-8b good for coding?
In our benchmarks, deepseek/deepseek-r1-0528-qwen3-8b scores 6.5/100 for coding. It runs at about 37.3 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 deepseek/deepseek-r1-0528-qwen3-8b good for agentic (tool-using) tasks?
In our benchmarks, deepseek/deepseek-r1-0528-qwen3-8b scores 45.1/100 for agentic workflows. It runs at about 37.3 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 deepseek-r1-0528-qwen3-8b for local inference?
Across 2 community benchmark runs, deepseek-r1-0528-qwen3-8b reaches up to 42.8 tok/s and averages 37.3 tok/s, with the fastest results on Apple M4 Max.
How much memory does deepseek-r1-0528-qwen3-8b need?
The leanest observed configuration used about 4.7 GB of memory.
Which tools have been used to run deepseek-r1-0528-qwen3-8b?
Benchmarks were submitted using LM Studio. Results are community-contributed and updated as new runs arrive.