Skip to content

Agents A1 VRAM requirements

How much VRAM Agents A1 needs
4-bit weights: 19.8 GBWith 8K context: 21.0 GBWith 32K context: 21.5 GB8-bit, 8K context: 35.9 GB
Parameters
35.1B
Active per token
3.9B
Longest context
256K
Released
Jun 2026
License
Apache 2.0
Strongest at
Not rated

How much VRAM Agents A1 needs

Memory Agents A1 needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit65.4 GB66.6 GB67.1 GB68.9 GB
8-bit34.7 GB35.9 GB36.4 GB38.3 GB
6-bit26.8 GB28.0 GB28.5 GB30.4 GB
5-bit23.3 GB24.5 GB24.9 GB26.8 GB
4-bit19.8 GB21.0 GB21.5 GB23.4 GB
3-bit16.0 GB17.2 GB17.7 GB19.5 GB
2-bit12.3 GB13.5 GB13.9 GB15.8 GB
1.6-bit6.7 GB7.9 GB8.4 GB10.3 GB

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

GPUs and Macs that run Agents A1

The best way Agents A1 runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Agents A1 on RTX 4060 report
RTX 5060Too bigPDF: Agents A1 on RTX 5060 report
RTX 3070Too bigPDF: Agents A1 on RTX 3070 report
RTX 3080 10 GBToo bigPDF: Agents A1 on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: Agents A1 on RTX 3060 12 GB report
Arc B580Too bigPDF: Agents A1 on Arc B580 report
RTX 4070Too bigPDF: Agents A1 on RTX 4070 report
RTX 4070 SuperToo bigPDF: Agents A1 on RTX 4070 Super report
RTX 5070Too bigPDF: Agents A1 on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: Agents A1 on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: Agents A1 on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: Agents A1 on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: Agents A1 on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: Agents A1 on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: Agents A1 on RTX 4080 Super report
RTX 5070 TiToo bigPDF: Agents A1 on RTX 5070 Ti report
RTX 5080Too bigPDF: Agents A1 on RTX 5080 report
Radeon RX 7900 XTX4-bit96K128 tok/s1.0 sPDF: Agents A1 on Radeon RX 7900 XTX report
RTX 30904-bit96K164 tok/s0.6 sPDF: Agents A1 on RTX 3090 report
RTX 40904-bit96K171 tok/s0.3 sPDF: Agents A1 on RTX 4090 report
Radeon AI PRO R97006-bit128K83 tok/s0.6 sPDF: Agents A1 on Radeon AI PRO R9700 report
RTX 50906-bit128K200 tok/s0.2 sPDF: Agents A1 on RTX 5090 report
2× RTX 30908-bit256K113 tok/s0.4 sPDF: Agents A1 on 2× RTX 3090 report
2× RTX 40908-bit256K120 tok/s0.2 sPDF: Agents A1 on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit256K145 tok/s0.3 sPDF: Agents A1 on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit256K115 tok/s0.2 sPDF: Agents A1 on RTX 6000 Ada report
L40S8-bit256K107 tok/s0.4 sPDF: Agents A1 on L40S report
2× RTX 50908-bit256K174 tok/s0.1 sPDF: Agents A1 on 2× RTX 5090 report
4× RTX 30908-bit256K113 tok/s0.2 sPDF: Agents A1 on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit256K174 tok/s0.2 sPDF: Agents A1 on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: Agents A1 on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: Agents A1 on Mac mini M4 24 GB report
Mac mini M4 32 GB3-bit176K39 tok/s14 sPDF: Agents A1 on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: Agents A1 on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB6-bit256K50 tok/s7.0 sPDF: Agents A1 on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit256K41 tok/s7.0 sPDF: Agents A1 on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit256K63 tok/s3.5 sPDF: Agents A1 on M4 Max 128 GB report
M5 Max 128 GB8-bit256K68 tok/s1.0 sPDF: Agents A1 on M5 Max 128 GB report
M3 Ultra 96 GB8-bit256K74 tok/s2.0 sPDF: Agents A1 on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit256K74 tok/s2.0 sPDF: Agents A1 on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit256K74 tok/s2.0 sPDF: Agents A1 on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit256K91 tok/s0.6 sPDF: Agents A1 on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit256K32 tok/s2.0 sPDF: Agents A1 on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit256K32 tok/s2.0 sPDF: Agents A1 on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit256K45 tok/s0.7 sPDF: Agents A1 on RTX Spark report
DGX Spark8-bit256K41 tok/s0.8 sPDF: Agents A1 on DGX Spark report
Dual DGX Spark8-bit256K41 tok/s0.4 sPDF: Agents A1 on Dual DGX Spark report
DGX Station GB3008-bit256K308 tok/s0.1 sPDF: Agents A1 on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit256K174 tok/s0.1 sPDF: Agents A1 on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit256K174 tok/s0.1 sPDF: Agents A1 on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit256K174 tok/s0.1 sPDF: Agents A1 on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit256K187 tok/s0.4 sPDF: Agents A1 on A100 80 GB SXM report
H100 SXM8-bit256K238 tok/s0.1 sPDF: Agents A1 on H100 SXM report
4× H200 NVL8-bit256K274 tok/s0.1 sPDF: Agents A1 on 4× H200 NVL report
8× H200 NVL8-bit256K274 tok/s0.1 sPDF: Agents A1 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 huihui-ai/Huihui-Agents-A1-abliterated-GGUF
vLLM
vllm serve InternScience/Agents-A1
MLX (Mac)
mlx_lm.chat --model mlx-community/Agents-A1-4bit

Questions

How much VRAM does Agents A1 need?

Agents A1 needs 19.8 GB for its weights at 4-bit, 21.0 GB with an 8K context and 21.5 GB with 32K. At 8-bit, 35.9 GB.

What is the minimum VRAM for Agents A1?

About 17.2 GB at 3-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 Radeon RX 7900 XTX.

How big is Agents A1?

35.1B parameters (3.9B active per token): 65.4 GB of weights at 16-bit, 19.8 GB at 4-bit.

Can Agents A1 run on an RTX 4090?

Yes. At 4-bit it holds 96K of context and writes about 171 tokens a second.

More models