WeiboAI · text

VibeThinker-3B

WeiboAI/VibeThinker-3B

VibeThinker-3B at Q4_K_M is exactly 1,929,902,176 bytes (1.80 GiB / 1.93 GB) — an effective 5.003 bits per weight, not the nominal 4. Its KV cache at 32K is 1.13 GiB.

From the file· summed from 1 file(s)From the file· KV per layer
Parameters
3.1B
Architecture
qwen2
36 layers
Context
131,072
native (config.json)
License
mit

Shipped quantizations

● exact bytes, summed from published files
QuantSize●Exact bytes●Effective bpw●Tensors●Publisher
Q2_K1.19 GiB1,274,755,1683.305—squ11z1
Q2_K1.19 GiB1,274,755,7763.305—prithivMLmods
Q3_K_S1.35 GiB1,454,356,5763.770—squ11z1
Q3_K_S1.35 GiB1,454,357,1843.770—prithivMLmods
Q3_K_M1.48 GiB1,590,474,8484.123434squ11z1
Q3_K_M1.48 GiB1,590,475,4564.123434prithivMLmods
Q3_K_L1.59 GiB1,707,391,0724.426—squ11z1
Q3_K_L1.59 GiB1,707,391,6804.426—prithivMLmods
Q4_01.70 GiB1,822,849,7284.726434prithivMLmods
Q4_K_S1.71 GiB1,834,383,4564.755—squ11z1
Q4_K_S1.71 GiB1,834,384,0644.755—prithivMLmods
Q4_K_M1.80 GiB1,929,902,1765.003434squ11z1
Q4_K_M1.80 GiB1,929,902,7845.003434semparuthiveeran
Q4_K_M1.80 GiB1,929,902,7845.003—prithivMLmods
Q5_K_S2.02 GiB2,169,665,6325.625—squ11z1
Q5_02.02 GiB2,169,666,2405.625—prithivMLmods
Q5_K_S2.02 GiB2,169,666,2405.625—prithivMLmods
Q5_K_M2.07 GiB2,224,814,1765.768434squ11z1
Q5_K_M2.07 GiB2,224,814,7845.768434prithivMLmods
Q6_K2.36 GiB2,538,158,1766.580434squ11z1
Q6_K2.36 GiB2,538,158,7846.580434prithivMLmods
Q8_03.06 GiB3,285,475,4248.517434squ11z1
Q8_03.06 GiB3,285,476,0328.517434prithivMLmods
F165.75 GiB6,178,316,38416.017434squ11z1
F165.75 GiB6,178,316,99216.017434prithivMLmods
BF165.75 GiB6,178,316,99216.017—prithivMLmods
F3211.50 GiB12,349,711,04032.015—prithivMLmods

KV cache by context

computed per layer
ContextKV cache (f16)●Flat formulaOverstated byFull / windowed / recurrent
4,0960.14 GiB0.14 GiB—36 / 0 / 0
8,1920.28 GiB0.28 GiB—36 / 0 / 0
16,3840.56 GiB0.56 GiB—36 / 0 / 0
32,7681.13 GiB1.13 GiB—36 / 0 / 0
65,5362.25 GiB2.25 GiB—36 / 0 / 0
131,0724.50 GiB4.50 GiB—36 / 0 / 0

Compare with

same modality, comparable size
vs Qwen3-4Bvs Qwen3-8Bvs HyperCLOVAX-SEED-Text-Instruct-1.5Bvs Llama-3.2-1B-Instructvs Qwen3-VL-8B-Instruct-abliterated-v1vs Llama-3.1-8B-InstructAll alternatives →Will it fit on my card? →

Will it run on your card?

full quant x context sweep
By VRAM:8GB12GB16GB24GB32GB48GB
Radeon RX 6500 XT 4GBGeForce RTX 3050 6GBGeForce RTX 5050 8GBGeForce RTX 3080 10GBGeForce RTX 2080 Ti 11GBGeForce RTX 5070 12GBGeForce RTX 5060 Ti 16GBApple M3 Pro 18GBGeForce RTX 3080 Ti 20GBGeForce RTX 5090 D V2 24GBGeForce RTX 5090 D 32GBApple M5 Max 36GBApple M5 Pro 48GBApple M5 Max 64GBApple M3 Ultra 96GBApple M5 Max 128GBApple M2 Ultra 192GBApple M3 Ultra 256GBApple M3 Ultra 512GB

Why other calculators give a different number

A parameters × bits ÷ 8 estimate puts Q4_K_M at roughly 1.62 GiB. The real file is 1.80 GiB, because a quantization is a mixture and some tensors are always kept at higher precision.

Architecture

from config.json
Layers
36
Attention heads
16
KV heads
2
Head dim
128
Hidden size
2048
Vocab
151,936
Sliding window
32768
SWA period
—
MLA
no
Experts
—
Experts per token
—
use_sliding_window
false

This model declares a sliding window but sets use_sliding_window: false, so the window is not applied. Honouring the field without the flag understates KV for the whole family.

Questions people ask

How much VRAM does VibeThinker-3B need?
Q4_K_M is exactly 1,929,902,176 bytes (1.80 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is VibeThinker-3B's KV cache?
1.13 GiB at 32K context with an f16 cache, computed per layer. Quantizing the cache to q8_0 roughly halves it, which is often the difference between a context length fitting and not.
Which quantization of VibeThinker-3B should I use?
Q4_K_M is the usual default. Pick the largest quantization that fits your card at the context you actually need — the table above gives exact sizes for every one published.