RunsOnMy ModelsGPUs & MacsAPI pricesVRAM calculatorNew

Can I run LightOnOCR 3 4B locally?

LightOnOCR 3 4B by lightonai has 4.5B parameters. At 4-bit with an 8K context it needs about 3.6 GB of GPU memory. It fits on 65 of 65 GPUs and Macs we track.

3.6 GBQ4_K_M · 8K context
RTX 4060smallest consumer GPU that fits
M1 16GBsmallest Mac that fits
256Kmax context
GQAattention type

Memory needed by quant and context length

GB of GPU memory = weights + KV cache (FP16) + runtime overhead. Weights use real file sizes from prithivMLmods/LightOnOCR-3-4B-GGUF where available (marked •), otherwise parameter count × bits per weight. How we calculate.

QuantQuality4K ctx8K ctx32K ctx128K ctx
2-bit (Q2_K)rough2.6 GB2.7 GB3.5 GB6.5 GB
3-bit (Q3_K_M) •usable3.1 GB3.2 GB4.0 GB7.0 GB
4-bit (Q4_K_M) •good, most popular3.5 GB3.6 GB4.4 GB7.4 GB
5-bit (Q5_K_M) •very good3.9 GB4.0 GB4.7 GB7.7 GB
6-bit (Q6_K) •near lossless4.2 GB4.3 GB5.1 GB8.1 GB
8-bit (Q8_0) •effectively lossless5.2 GB5.3 GB6.0 GB9.0 GB
16-bit (F16/BF16) •original weights8.8 GB9.0 GB9.7 GB12.7 GB

Which GPUs and Macs run LightOnOCR 3 4B?

Best quality quant that fits at 8K context. "Tight" = within 10% of usable memory. Speed is a ceiling from memory bandwidth (real speed is typically 50–80% of it).

HardwareMemoryFits?Best quantNeedsMax tok/s (est.)
RTX 3060 12GB12 GBFitsF169.0 GB43details
RTX 4060 Ti 16GB16 GBFitsF169.0 GB34details
RTX 309024 GBFitsF169.0 GB111details
RTX 409024 GBFitsF169.0 GB120details
RTX 509032 GBFitsF169.0 GB213details
RTX 5070 Ti16 GBFitsF169.0 GB106details
RX 7900 XTX24 GBFitsF169.0 GB114details
Mac M4 Pro 48GB48 GBFitsF169.0 GB32details
Mac M4 Max 128GB128 GBFitsF169.0 GB65details
Mac M3 Ultra 512GB512 GBFitsF169.0 GB97details
DGX Spark128 GBFitsF169.0 GB32details
RTX 407012 GBFitsF169.0 GB60details
RTX 5060 Ti 16GB16 GBFitsF169.0 GB53details
RTX 40608 GBFitsQ8_05.3 GB61details
Arc B58012 GBFitsF169.0 GB54details
RTX 3060 Ti8 GBFitsQ8_05.3 GB100details
RTX 30708 GBFitsQ8_05.3 GB100details
RTX 3080 10GB10 GBFitsQ8_05.3 GB170details
RTX 3080 12GB12 GBFitsF169.0 GB108details
RTX 3090 Ti24 GBFitsF169.0 GB120details
RTX 4060 Ti 8GB8 GBFitsQ8_05.3 GB64details
RTX 4070 Super12 GBFitsF169.0 GB60details
RTX 4070 Ti12 GBFitsF169.0 GB60details
RTX 4070 Ti Super16 GBFitsF169.0 GB80details
RTX 408016 GBFitsF169.0 GB85details
RTX 4080 Super16 GBFitsF169.0 GB87details
RTX 50608 GBFitsQ8_05.3 GB100details
RTX 507012 GBFitsF169.0 GB80details
RTX 508016 GBFitsF169.0 GB114details
2x RTX 30902×24 GBFitsF169.0 GB111details
2x RTX 40902×24 GBFitsF169.0 GB120details
2x RTX 50902×32 GBFitsF169.0 GB213details
RTX A600048 GBFitsF169.0 GB91details
RTX 6000 Ada48 GBFitsF169.0 GB114details
RTX PRO 6000 Blackwell96 GBFitsF169.0 GB213details
L40S48 GBFitsF169.0 GB103details
A100 40GB40 GBFitsF169.0 GB185details
A100 80GB80 GBFitsF169.0 GB242details
H100 PCIe80 GBFitsF169.0 GB237details
H100 SXM80 GBFitsF169.0 GB398details
H200141 GBFitsF169.0 GB570details
RX 76008 GBFitsQ8_05.3 GB64details
RX 7700 XT12 GBFitsF169.0 GB51details
RX 7800 XT16 GBFitsF169.0 GB74details
RX 7900 GRE16 GBFitsF169.0 GB68details
RX 7900 XT20 GBFitsF169.0 GB95details
RX 907016 GBFitsF169.0 GB76details
RX 9070 XT16 GBFitsF169.0 GB76details
Ryzen AI Max+ 395 128GB128 GBFitsF169.0 GB30details
Arc A770 16GB16 GBFitsF169.0 GB66details
Mac M1 16GB16 GBFitsF169.0 GB8.1details
Mac M2 24GB24 GBFitsF169.0 GB12details
Mac M3 24GB24 GBFitsF169.0 GB12details
Mac M4 16GB16 GBFitsF169.0 GB14details
Mac M4 32GB32 GBFitsF169.0 GB14details
Mac M1 Pro 32GB32 GBFitsF169.0 GB24details
Mac M3 Pro 36GB36 GBFitsF169.0 GB18details
Mac M4 Pro 24GB24 GBFitsF169.0 GB32details
Mac M4 Pro 64GB64 GBFitsF169.0 GB32details
Mac M1 Max 64GB64 GBFitsF169.0 GB47details
Mac M2 Max 96GB96 GBFitsF169.0 GB47details
Mac M3 Max 128GB128 GBFitsF169.0 GB47details
Mac M4 Max 64GB64 GBFitsF169.0 GB65details
Mac M2 Ultra 192GB192 GBFitsF169.0 GB95details
Mac M3 Ultra 256GB256 GBFitsF169.0 GB97details

Questions

How much VRAM does LightOnOCR 3 4B need?

About 3.6 GB at 4-bit (Q4_K_M) with an 8K context: 2.5 GB of weights (real GGUF file size), 0.3 GB of KV cache and 0.9 GB of runtime overhead. At 8-bit it needs about 5.3 GB.

What is the smallest GPU that runs LightOnOCR 3 4B?

At Q4_K_M and 8K context, the smallest single consumer GPU in our table that fits it is the RTX 4060 (8 GB).

Can I run LightOnOCR 3 4B on a Mac?

Yes. At Q4_K_M with 8K context it fits on a Mac M1 16GB or larger, using the default macOS GPU memory limit.

Model facts

Hugging Face repolightonai/LightOnOCR-3-4B
Parameters4,539,265,536 (4.5B)
Layers / hidden size32 / 2560
AttentionGQA (hybrid: 24 linear/state layers); 16 heads, 4 KV heads, head dim 256
KV cache per token (FP16)32.0 KiB
Max context262,144 tokens
Licenseapache-2.0
Released on HFOct 6, 2026
Downloads (30d) / likes14,321 / 146

Sources: Hugging Face model API · config.json (lightonai/LightOnOCR-3-4B) · GGUF files · GPU specs from vendor pages (list). Data fetched Nov 17, 2026. Spot an error? Report it.