Model benchmarks

spark-2.6-4b local LLM performance

As of November 2026, spark-2.6-4b runs at up to 144.7 tok/s for local inference (best of 2 community benchmark runs across 1 GPU).

OpenAI-compatible
ShareRedditX

Model size

4B

Peak speed

144.7 tok/s

Average speed

144.3 tok/s

Avg PP

4216.5 tok/s

Min memory

4.2 GB

Max context

32,768 tokens

Avg output / run

6,750 tokens

Avg runtime / run

59s

Avg quality

54.0

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)
Overall50.354.057.6
Agent Workflow75.677.980.1
Code Generation38.442.346.1
Role Play & Narrative57.059.762.4
Research & Analysis25.736.146.4

Performance by hardware and tool

Every hardware/tool/quantization combination spark-2.6-4b has been benchmarked on, ranked by peak token generation speed. Last updated November 2026.

HardwareToolQuantBest tok/sAvg tok/sMemoryContextQualityRuns
NVIDIA GeForce RTX 3080 TiOpenAI-compatible—144.7 tok/s144.3 tok/s4.2 GB32,768 tokens54.02

Benchmark runs

All 2 spark-2.6-4b runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.

Frequently asked questions

Is spark-2.6-4b good for agentic (tool-using) tasks?
In our benchmarks, spark-2.6-4b scores 77.9/100 for agentic workflows. It runs at about 144.3 tok/s, so if you want more speed, unsloth/Qwen3.6-35B-A3B-GGUF:UD-Q4_K_M is faster (~246.8 tok/s) and still scores well for agentic workflows (80.6/100).
How fast is spark-2.6-4b for local inference?
Across 2 community benchmark runs, spark-2.6-4b reaches up to 144.7 tok/s and averages 144.3 tok/s, with the fastest results on NVIDIA GeForce RTX 3080 Ti.
How much memory does spark-2.6-4b need?
The leanest observed configuration used about 4.2 GB of memory.
Which tools have been used to run spark-2.6-4b?
Benchmarks were submitted using OpenAI-compatible. Results are community-contributed and updated as new runs arrive.