Skip to content

Inkling VRAM requirements

How much VRAM Inkling needs
4-bit weights: 538 GBWith 4K context: 540 GB8-bit, 4K context: 944 GB
Parameters
952B
Active per token
40.9B
Longest context
4K
Released
Jul 2026
License
Apache 2.0
Strongest at
Text and images

How much VRAM Inkling needs

Memory Inkling needs at each precision and context length
PrecisionWeightsWith 4K
16-bit1,774 GB1,776 GB
8-bit942 GB944 GB
6-bit727 GB729 GB
5-bit631 GB633 GB
4-bit538 GB540 GB
3-bit434 GB436 GB
2-bit333 GB335 GB
1.6-bit183 GB185 GB

One conversation, 16-bit context cache, runtime overhead included.

GPUs and Macs that run Inkling

The best way Inkling runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Inkling on RTX 4060 report
RTX 5060Too bigPDF: Inkling on RTX 5060 report
RTX 3070Too bigPDF: Inkling on RTX 3070 report
RTX 3080 10 GBToo bigPDF: Inkling on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: Inkling on RTX 3060 12 GB report
Arc B580Too bigPDF: Inkling on Arc B580 report
RTX 4070Too bigPDF: Inkling on RTX 4070 report
RTX 4070 SuperToo bigPDF: Inkling on RTX 4070 Super report
RTX 5070Too bigPDF: Inkling on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: Inkling on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: Inkling on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: Inkling on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: Inkling on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: Inkling on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: Inkling on RTX 4080 Super report
RTX 5070 TiToo bigPDF: Inkling on RTX 5070 Ti report
RTX 5080Too bigPDF: Inkling on RTX 5080 report
Radeon RX 7900 XTXToo bigPDF: Inkling on Radeon RX 7900 XTX report
RTX 3090Too bigPDF: Inkling on RTX 3090 report
RTX 4090Too bigPDF: Inkling on RTX 4090 report
Radeon AI PRO R9700Too bigPDF: Inkling on Radeon AI PRO R9700 report
RTX 5090Too bigPDF: Inkling on RTX 5090 report
2× RTX 3090Too bigPDF: Inkling on 2× RTX 3090 report
2× RTX 4090Too bigPDF: Inkling on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GBToo bigPDF: Inkling on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 AdaToo bigPDF: Inkling on RTX 6000 Ada report
L40SToo bigPDF: Inkling on L40S report
2× RTX 5090Too bigPDF: Inkling on 2× RTX 5090 report
4× RTX 3090Too bigPDF: Inkling on 4× RTX 3090 report
RTX PRO 6000 BlackwellToo bigPDF: Inkling on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: Inkling on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: Inkling on Mac mini M4 24 GB report
Mac mini M4 32 GBToo bigPDF: Inkling on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: Inkling on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GBToo bigPDF: Inkling on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GBToo bigPDF: Inkling on Mac mini M4 Pro 64 GB report
M4 Max 128 GBToo bigPDF: Inkling on M4 Max 128 GB report
M5 Max 128 GBToo bigPDF: Inkling on M5 Max 128 GB report
M3 Ultra 96 GBToo bigPDF: Inkling on M3 Ultra 96 GB report
M3 Ultra 256 GB1.6-bit4K40 tok/s25 sPDF: Inkling on M3 Ultra 256 GB report
M3 Ultra 512 GB2-bit4K26 tok/s25 sPDF: Inkling on M3 Ultra 512 GB report
M5 Ultra 512 GB2-bit4K35 tok/s7.9 sPDF: Inkling on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GBToo bigPDF: Inkling on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GBToo bigPDF: Inkling on Ryzen AI Max+ 395 128 GB report
RTX SparkToo bigPDF: Inkling on RTX Spark report
DGX SparkToo bigPDF: Inkling on DGX Spark report
Dual DGX Spark1.6-bit4K21 tok/s5.2 sPDF: Inkling on Dual DGX Spark report
DGX Station GB3001.6-bit4K177 tok/s0.7 sPDF: Inkling on DGX Station GB300 report
2× RTX PRO 6000 Blackwell1.6-bit4K94 tok/s1.2 sPDF: Inkling on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell2-bit4K62 tok/s0.7 sPDF: Inkling on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell6-bit4K33 tok/s0.4 sPDF: Inkling on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXMToo bigPDF: Inkling on A100 80 GB SXM report
H100 SXMToo bigPDF: Inkling on H100 SXM report
4× H200 NVL4-bit4K88 tok/s0.7 sPDF: Inkling on 4× H200 NVL report
8× H200 NVL8-bit4K59 tok/s0.3 sPDF: Inkling on 8× H200 NVL report

Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.

How to run it

llama.cpp
llama-server -hf unsloth/inkling-GGUF
vLLM
vllm serve thinkingmachines/Inkling

Questions

How much VRAM does Inkling need?

Inkling needs 538 GB for its weights at 4-bit, 540 GB with an 4K context and 540 GB with 4K. At 8-bit, 944 GB.

What is the minimum VRAM for Inkling?

About 185 GB at 1.6-bit with an 4K context, the lowest precision we recommend for a model this size. The smallest machine here that runs it without spilling is the M3 Ultra 256 GB.

How big is Inkling?

952B parameters (40.9B active per token): 1,774 GB of weights at 16-bit, 538 GB at 4-bit.

Can Inkling run on an RTX 4090?

Not on one card. The smallest machine here that runs it is the M3 Ultra 256 GB.

More models