Laguna-S-2.1

Laguna family · 118.0B params

3recorded runs

125.5best tok/s

2verified

Evidence: 2 sourced runs · 1 awaiting review · 1 independent source · first recorded Aug 1, 2026 · last recorded Sep 4, 2026. What these labels mean.

What the Laguna-S-2.1 numbers mean

The fastest verified result on our board is 125.5 output tok/s on an Intel Arc Pro B70 x4 at INT4_W4A16 with a 8192-token context. That is about 8.0 ms per generated token — 500 tokens takes roughly 4.0 seconds. Across the 3 records we hold for this model the range is 23.9–125.5 tok/s on 3 GPUs, because quantisation, context length and the engine build all move the number.

How much VRAM does Laguna-S-2.1 need?

Laguna-S-2.1 has 118.0B parameters. These estimates use the same formula as the VRAM calculator: weights, plus about 15% for a moderate-context KV cache and runtime overhead, plus a small engine floor.

Quantisation	Estimated VRAM	Smallest card that fits
Q4_K_M (4-bit)	77.1 GB	96 GB
Q8_0 (8-bit)	145.0 GB	192 GB
FP16 (16-bit)	272.2 GB	multi-GPU or CPU offload

Long contexts and large batches raise the real number — read the smallest card as "possible with a reduced context". Measured throughput is on the benchmark board.

Fastest GPUs for this model (best tok/s)

Models with a similar footprint

Model	Params	Q4_K_M weights
gpt-oss-120b	117B	65.8 GB
NVIDIA-Nemotron-3.5-Super-VL-120B-A12B	120B	67.5 GB
Qwen 3.8 Flash	125B	70.3 GB
Qwen 3.8 Flash-Next Base	125B	70.3 GB
Ling-3.0-flash	127B	71.4 GB
Llama-4-Scout-17B-16E-Instruct	109B	61.3 GB

Ranked by distance from Laguna-S-2.1's parameter count. Full benchmark board.

All benchmarks Methodology

GPU	Model	Quant	Backend	Ctx	tok/s	prompt tok/s	VRAM GB	Source
Intel Arc Pro B70 x4 · 32 GB	Laguna-S-2.1	INT4_W4A16	vllm 0.1.dev1172+g4a6fd8747; local vLLM 1…	8192	125.5 ✓	—	—	src
CMP 170HX x2 · 64 GB	Laguna-S-2.1	Q4_K_M	llama.cpp 0.28.0	32768	63.3 ✓	1118.5	74.6	src
Multi-GPU x3 · 72 GB	Laguna-S-2.1	UD-IQ4_XS	llama.cpp 81 (15586e2)	278528	23.9 …	—	65.1	src