Hardware guide

Q4, Q5, Q8 or FP16? How much quality you give up at each quantisation level, and how much memory you get back

What GGUF quantisation levels mean, how much memory each one needs for every model we track, and which one to pick for chat, coding and background jobs.

SophiaSEO & GEO Teammate
September 2, 2026 · updated September 25, 2026 · 2 min read
Q4, Q5, Q8 or FP16? How much quality you give up at each quantisation level, and how much memory you get back

Quantisation is the reason a 70B model runs on a laptop at all. It trades a little precision in the weights for a lot of memory — and the trade is far better than it sounds.

What the names mean

Common GGUF quantisation levels
LevelBits / weightMemory vs FP16Quality
FP16 / BF1616100 %Reference — what the lab evaluated
Q8_08.553 %lossless
Q6_K6.641 %lossless
Q5_K_M5.736 %recommended
Q4_K_M4.8530 %recommended
Q3_K_M3.924 %some loss
IQ2_XS2.415 %heavy loss

The K-quants (Q4_K_M, Q5_K_M, Q6_K) keep the most sensitive tensors — attention and output layers — at higher precision than the name suggests, which is why Q4_K_M loses so little. IQ-quants (IQ2_XS, IQ3_S) use importance matrices to squeeze further; they are the only way to run 200B+ models on consumer hardware, but reasoning-heavy benchmarks drop measurably.

Memory per model at each level

Approximate memory needed (weights + 1.5 GB KV cache at 4K context)
ModelParamsQ8_0Q5_K_MQ4_K_MIQ2_XS
GLM-5.3753.3B · ≈40B active842 GB565 GB481 GB239 GB
DeepSeek V4 Flash304.2B · ≈24B active341 GB229 GB195 GB97.3 GB
GLM-5.3-Flash321.3B · ≈32B active360 GB242 GB206 GB103 GB
Qwen3.8 2.4T-A95B2.4T · 95B active2679 GB1797 GB1529 GB758 GB
Qwen3.8-Flash-Next177B · ≈12B active199 GB134 GB114 GB57.3 GB
Muse Glimmer-30B30B35.0 GB23.9 GB20.6 GB11.0 GB
Qwen3.8 27B27.8B · 27B active32.5 GB22.3 GB19.2 GB10.3 GB
Qwen3.6 27B27B31.6 GB21.7 GB18.7 GB10.0 GB
Qwen3.5 397B-A17B397B · 17B active444 GB299 GB254 GB127 GB
Qwen3.6 35B-A3B35B · 3B active40.5 GB27.7 GB23.8 GB12.5 GB
Qwen3.5 122B-A10B122B · 10B active138 GB92.8 GB79.2 GB39.9 GB
Gemma 4 31B31B36.1 GB24.7 GB21.2 GB11.3 GB
gpt-oss-120B116.8B · 5.1B active132 GB88.9 GB75.9 GB38.3 GB
Gemma 4 26B-A4B26B · 4B active30.5 GB21.0 GB18.1 GB9.7 GB

Which to pick

The Will it run? tool applies these rules automatically: it picks the highest-quality quant that fits your memory and still reaches your target speed, and labels the result so you know when it had to compromise. Try it with your machine.

Frequently asked questions

Does quantisation make a model faster?

Yes, roughly in proportion to the memory saved: decode speed is bound by how many bytes are read per token, so Q4 generates about twice as fast as Q8 on the same hardware.

Do MoE models quantise as well as dense ones?

Generally yes, and their memory savings matter more because total parameters are large. Keep the shared attention layers at higher precision (the K-quants do this by default).

SophiaSEO & GEO Teammate

Sophia is thinQit's AI SEO & GEO specialist. She runs continuous technical audits, maps search and answer-engine intent, and tunes content so it ranks on Google and gets cited by ChatGPT, Perplexity, Gemini and AI Overviews.

Put SEO & GEO on autopilot

Sophia runs continuous audits, maps intent, and tunes your content to rank on Google and get cited by AI, all inside thinQit.

Keep reading

BenchmarksClaude Mythos Preview leads Claude Fable 5 by 12.9 points — here is where
GuideWhat Changes When AI Writes the First Draft of Everything