Anthropic

Claude 4.5 Haiku (70B)

Fast model tuned for low-latency responses

Model Summary

Family

Claude 4.5

Version

4.5

Parameters

70B (est.)

Parameter counts for closed models are estimates; vendors rarely publish exact sizes.

VRAM Requirements by Quantization

Memory needed to serve Claude 4.5 Haiku (70B) for inference, including a 20% overhead for activations and KV cache.

PrecisionVRAM neededSmallest single GPU that fits
INT4 (4-bit)39.12 GBNVIDIA L40S (48 GB)
INT8 (8-bit)78.23 GBNVIDIA H100 SXM5 (80 GB)
FP16 (16-bit)156.46 GBNVIDIA B200 SXM (192 GB)
FP32 (32-bit)312.92 GBMulti-GPU required

Recommended GPU Configurations

Cheapest on-demand configurations to serve Claude 4.5 Haiku (70B) at 8-bit (78 GB VRAM).

1x NVIDIA A100 80GB

80 GB total VRAM · Ampere

~$4.50/h

1x AMD Instinct MI300X

192 GB total VRAM · CDNA 3

~$6.00/h

2x NVIDIA L40S

96 GB total VRAM · Ada Lovelace

~$7.00/h

Quick GPU Planning

Use the calculator pre-filled with this exact version to estimate memory, speed, and compute requirements in a few clicks.

Access Pre-filled Calculator

Frequently Asked Questions

How much VRAM do you need to run Claude 4.5 Haiku (70B)?

With an estimated 70B parameters, Claude 4.5 Haiku (70B) needs roughly 78 GB of VRAM in 8-bit (INT8), 39 GB in 4-bit, and 156 GB in FP16, including a 20% overhead for activations and KV cache.

Which GPUs can run Claude 4.5 Haiku (70B)?

At 8-bit quantization, the most cost-effective option is 1x NVIDIA A100 80GB (80 GB combined VRAM, around $4.50/hour on-demand). Higher-end cards like the NVIDIA B200 or AMD MI355X reduce the GPU count needed.

Can Claude 4.5 Haiku (70B) run on a single GPU?

Yes. In 8-bit, a single NVIDIA H100 SXM5 (80 GB) fits the model.

How much does it cost to serve Claude 4.5 Haiku (70B) in the cloud?

Renting 1x A100 80GB costs on the order of $4.50/hour, i.e. about $3,285/month running 24/7. Actual prices vary by provider and commitment; spot and reserved capacity can be significantly cheaper.