Model benchmarks

KAT-Coder-V2.5-Dev.i1-IQ3_XXS local LLM performance

As of September 2026, KAT-Coder-V2.5-Dev.i1-IQ3_XXS runs at up to 86.9 tok/s for local inference (best of 2 community benchmark runs across 1 GPU).

llama.cpp
ShareRedditX

Model size

Unknown

Peak speed

86.9 tok/s

Average speed

85.7 tok/s

Avg PP

1198.1 tok/s

Min memory

15.6 GB

Max context

131,072 tokens

Avg output / run

9,227 tokens

Avg runtime / run

1m 54s

Avg quality

76.8

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)
Overall76.676.877.1
Agent Workflow81.382.082.8
Code Generation69.173.778.3
Role Play & Narrative71.071.471.9
Research & Analysis77.780.182.5

Performance by hardware and tool

Every hardware/tool/quantization combination KAT-Coder-V2.5-Dev.i1-IQ3_XXS has been benchmarked on, ranked by peak token generation speed. Last updated September 2026.

HardwareToolQuantBest tok/sAvg tok/sMemoryContextQualityRuns
NVIDIA GeForce RTX 5060llama.cpp—86.9 tok/s85.7 tok/s15.6 GB131,072 tokens76.82

Benchmark runs

All 2 KAT-Coder-V2.5-Dev.i1-IQ3_XXS runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.

Frequently asked questions

Is KAT-Coder-V2.5-Dev.i1-IQ3_XXS good for coding?
In our benchmarks, KAT-Coder-V2.5-Dev.i1-IQ3_XXS scores 73.7/100 for coding. It runs at about 85.7 tok/s, so if you want more speed, Tiel-Coder-35B-A3B-MLX-oQ4e is faster (~107.9 tok/s) and still scores well for coding (81.5/100).
Is KAT-Coder-V2.5-Dev.i1-IQ3_XXS good for agentic (tool-using) tasks?
In our benchmarks, KAT-Coder-V2.5-Dev.i1-IQ3_XXS scores 82.0/100 for agentic workflows. It runs at about 85.7 tok/s, so if you want more speed, Tiel-Coder-35B-A3B-MLX-oQ4e-MTP is faster (~123.7 tok/s) and still scores well for agentic workflows (84.4/100).
How fast is KAT-Coder-V2.5-Dev.i1-IQ3_XXS for local inference?
Across 2 community benchmark runs, KAT-Coder-V2.5-Dev.i1-IQ3_XXS reaches up to 86.9 tok/s and averages 85.7 tok/s, with the fastest results on NVIDIA GeForce RTX 5060.
How much memory does KAT-Coder-V2.5-Dev.i1-IQ3_XXS need?
The leanest observed configuration used about 15.6 GB of memory.
Which tools have been used to run KAT-Coder-V2.5-Dev.i1-IQ3_XXS?
Benchmarks were submitted using llama.cpp. Results are community-contributed and updated as new runs arrive.