Skip to content

Beam 1.0 501B A23B VRAM requirements

How much VRAM Beam 1.0 501B A23B needs
4-bit weights: 283 GBWith 8K context: 284 GBWith 32K context: 285 GB8-bit, 8K context: 497 GB
Parameters
501B
Active per token
23B
Longest context
1M
Released
Oct 2026
License
Apache 2.0
Strongest at
Not rated

How much VRAM Beam 1.0 501B A23B needs

Memory Beam 1.0 501B A23B needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit934 GB935 GB936 GB939 GB
8-bit496 GB497 GB498 GB501 GB
6-bit383 GB384 GB385 GB388 GB
5-bit332 GB333 GB334 GB338 GB
4-bit283 GB284 GB285 GB289 GB
3-bit228 GB229 GB230 GB234 GB
2-bit175 GB176 GB177 GB181 GB
1.6-bit96.3 GB97.5 GB98.4 GB102 GB

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

GPUs and Macs that run Beam 1.0 501B A23B

The best way Beam 1.0 501B A23B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Beam 1.0 501B A23B on RTX 4060 report
RTX 5060Too bigPDF: Beam 1.0 501B A23B on RTX 5060 report
RTX 3070Too bigPDF: Beam 1.0 501B A23B on RTX 3070 report
RTX 3080 10 GBToo bigPDF: Beam 1.0 501B A23B on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: Beam 1.0 501B A23B on RTX 3060 12 GB report
Arc B580Too bigPDF: Beam 1.0 501B A23B on Arc B580 report
RTX 4070Too bigPDF: Beam 1.0 501B A23B on RTX 4070 report
RTX 4070 SuperToo bigPDF: Beam 1.0 501B A23B on RTX 4070 Super report
RTX 5070Too bigPDF: Beam 1.0 501B A23B on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: Beam 1.0 501B A23B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: Beam 1.0 501B A23B on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: Beam 1.0 501B A23B on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: Beam 1.0 501B A23B on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: Beam 1.0 501B A23B on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: Beam 1.0 501B A23B on RTX 4080 Super report
RTX 5070 TiToo bigPDF: Beam 1.0 501B A23B on RTX 5070 Ti report
RTX 5080Too bigPDF: Beam 1.0 501B A23B on RTX 5080 report
Radeon RX 7900 XTXToo bigPDF: Beam 1.0 501B A23B on Radeon RX 7900 XTX report
RTX 3090Too bigPDF: Beam 1.0 501B A23B on RTX 3090 report
RTX 4090Too bigPDF: Beam 1.0 501B A23B on RTX 4090 report
Radeon AI PRO R9700Too bigPDF: Beam 1.0 501B A23B on Radeon AI PRO R9700 report
RTX 5090Too bigPDF: Beam 1.0 501B A23B on RTX 5090 report
2× RTX 3090Too bigPDF: Beam 1.0 501B A23B on 2× RTX 3090 report
2× RTX 4090Too bigPDF: Beam 1.0 501B A23B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GBToo bigPDF: Beam 1.0 501B A23B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 AdaToo bigPDF: Beam 1.0 501B A23B on RTX 6000 Ada report
L40SToo bigPDF: Beam 1.0 501B A23B on L40S report
2× RTX 5090Too bigPDF: Beam 1.0 501B A23B on 2× RTX 5090 report
4× RTX 3090Too bigPDF: Beam 1.0 501B A23B on 4× RTX 3090 report
RTX PRO 6000 BlackwellToo bigPDF: Beam 1.0 501B A23B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: Beam 1.0 501B A23B on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: Beam 1.0 501B A23B on Mac mini M4 24 GB report
Mac mini M4 32 GBToo bigPDF: Beam 1.0 501B A23B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: Beam 1.0 501B A23B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GBToo bigPDF: Beam 1.0 501B A23B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GBToo bigPDF: Beam 1.0 501B A23B on Mac mini M4 Pro 64 GB report
M4 Max 128 GBToo bigPDF: Beam 1.0 501B A23B on M4 Max 128 GB report
M5 Max 128 GBToo bigPDF: Beam 1.0 501B A23B on M5 Max 128 GB report
M3 Ultra 96 GBToo bigPDF: Beam 1.0 501B A23B on M3 Ultra 96 GB report
M3 Ultra 256 GB2-bit184K43 tok/s14 sPDF: Beam 1.0 501B A23B on M3 Ultra 256 GB report
M3 Ultra 512 GB5-bit896K26 tok/s14 sPDF: Beam 1.0 501B A23B on M3 Ultra 512 GB report
M5 Ultra 512 GB5-bit896K35 tok/s4.4 sPDF: Beam 1.0 501B A23B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GBToo bigPDF: Beam 1.0 501B A23B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GBToo bigPDF: Beam 1.0 501B A23B on Ryzen AI Max+ 395 128 GB report
RTX Spark1.6-bit480K39 tok/s5.0 sPDF: Beam 1.0 501B A23B on RTX Spark report
DGX Spark1.6-bit480K36 tok/s5.0 sPDF: Beam 1.0 501B A23B on DGX Spark report
Dual DGX Spark2-bit1M21 tok/s2.9 sPDF: Beam 1.0 501B A23B on Dual DGX Spark report
DGX Station GB3003-bit288K186 tok/s0.4 sPDF: Beam 1.0 501B A23B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell2-bit160K100 tok/s0.7 sPDF: Beam 1.0 501B A23B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell5-bit832K62 tok/s0.4 sPDF: Beam 1.0 501B A23B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit1M44 tok/s0.3 sPDF: Beam 1.0 501B A23B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXMToo bigPDF: Beam 1.0 501B A23B on A100 80 GB SXM report
H100 SXMToo bigPDF: Beam 1.0 501B A23B on H100 SXM report
4× H200 NVL8-bit1M96 tok/s0.4 sPDF: Beam 1.0 501B A23B on 4× H200 NVL report
8× H200 NVL8-bit1M96 tok/s0.2 sPDF: Beam 1.0 501B A23B 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 bartowski/Beam-1.0-501B-A23B-GGUF
vLLM
vllm serve reflection/Beam-1.0-501B-A23B

Questions

How much VRAM does Beam 1.0 501B A23B need?

Beam 1.0 501B A23B needs 283 GB for its weights at 4-bit, 284 GB with an 8K context and 285 GB with 32K. At 8-bit, 497 GB.

What is the minimum VRAM for Beam 1.0 501B A23B?

About 97.5 GB at 1.6-bit with an 8K 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 Beam 1.0 501B A23B?

501B parameters (23B active per token): 934 GB of weights at 16-bit, 283 GB at 4-bit.

Can Beam 1.0 501B A23B run on an RTX 4090?

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

More models