models/qwen-qwen2.5-coder-32b-instruct

Qwen2.5 Coder 32B

Sources aider, arena, livebench · Confidence A

params

32B

dense

bench sources

high conf

context

—

max tokens

license

Apache 2.0

permissive

▸Model meta

LiveBench

instruction following64.9

coding57.0

language24.6

overall48.8

Aider · whole

polyglot pass rate16.4$0.00 run

▸Hardware fit matrix

by gpu tier	Q2_K 12.0 GB	Q4_K_M 16.0 GB	Q5_K_M 20.0 GB	Q6_K 24.0 GB	Q8_0 32.0 GB	FP16 64.0 GB
8 GB tier rtx 3050 · 4060
12 GB tier rtx 3060 · 4070
16 GB tier rtx 4080 · 4060 ti 16g	GREAT 46 t/s
24 GB tier rtx 3090 · 4090	GREAT 74 t/s	GREAT 56 t/s	OK 45 t/s
32 GB tier rtx 5090 · m3 max	GREAT 137 t/s	GREAT 104 t/s	GREAT 84 t/s	GREAT 70 t/s
48 GB tier a6000 · m3 max 64	GREAT 44 t/s	GREAT 34 t/s	GREAT 27 t/s	GREAT 23 t/s	GREAT 17 t/s
80 GB tier h100 · m3 ultra 128	GREAT 150 t/s	GREAT 114 t/s	GREAT 92 t/s	GREAT 77 t/s	GREAT 58 t/s	GREAT 29 t/s

full fit · production speed tight fit · usabledoesn't fit

Get personalized ranking

Tell us your machine — we'll tell you if this is actually your best pick, or what's better.