Only with CPU offload

Ling 3.0 Flash 124B-A5B at Q4_K_M needs 70.4 GB but only 10.6 GB is addressable, so about 86% of the layers would stream from system RAM at roughly 60 GB/s. Expect around 5.6 tokens per second — usable for batch work, painful for chat.

14% on GPU 8K context Q4_K_M · 69.7 GB MIT Released 2 Aug 2026 Not in the Ollama library

A 124B hybrid (5 linear-attention layers per MLA layer) with 5.1B active: SWE-bench Pro 56.6 and AIME 93 claimed. Built for 96–128 GB machines.

Hardware requirements What hardware do I need to run Ling 3.0 Flash 124B-A5B? →

At Q4_K_M with an 8K context: a 141 GB GPU (H200 SXM) or a 128 GB Mac. Minimum, recommended, high-quality and long-context tiers, with all 114 devices checked at every quantisation.

Fits instead: Ling 3.0 Tiny 7.9B-A1.3B (5.1 GB)

The VRAM budget

weights 69.7 GB
Weights 69.7 GB KV cache @ 8K 0.06 GB Runtime overhead 0.6 GB Over budget 59.8 GB past 10.6 GB

Quantisation ladder

QuantWeightsTotal @ 8KMax contextTok/sQualityFit
Q8_0 122.7 GB 123.4 GB — ~3.0 −0.1% ppl 112.8 GB over
Q6_K 94.7 GB 95.4 GB — ~3.9 −0.4% ppl 84.8 GB over
Q5_K_M 81.8 GB 82.5 GB — ~4.6 −0.8% ppl 71.9 GB over
Q4_K_M 69.7 GB 70.4 GB — ~5.6 −1.9% ppl 59.8 GB over
Q3_K_M 56.4 GB 57.1 GB — ~7.1 −5.4% ppl 46.5 GB over
Q2_K 48.4 GB 49.0 GB — ~8.5 −15% ppl 38.4 GB over

Quality is the published perplexity delta against f16 weights. Max context assumes an f16 KV cache; q8_0 roughly doubles it. Only 7 of its 42 blocks keep a per-token KV cache; the rest are linear-attention, Mamba or convolution blocks with a fixed-size state. This model uses multi-head latent attention, so its cache is a compressed latent rather than full K and V.

How to run it

terminal
$ llama-server \
    -hf inclusionAI/Ling-3.0-flash:Q4_K_M \
    -c 8192 -ngl 5

The engine underneath most of the others. Every knob is exposed. More on llama.cpp.

01Download is 69.7 GB. Keep it on an SSD — a first load off a spinning disk takes minutes.
02Close anything else holding VRAM. A browser with hardware acceleration can sit on 1–2 GB.
03If it falls back to CPU silently, drop the context first, then step down a quantisation.
See all models for this rig Compare with another model

Related models on an Arc B580

ModelTotalTok/s est.
Ling 3.0 Tiny 7.9B-A1.3B
7.9B MoE · same family
Q4_K_M · 5.1 GB 145 Runs great
Nemotron 3 Super 120B-A12B
124B MoE · similar size
Q4_K_M · 70.4 GB ~2.4 CPU offload
Qwen3.5 122B-A10B
125B MoE · similar size
Q4_K_M · 71.1 GB ~2.8 CPU offload
Mistral Small 4 119B-A6B
119B MoE · similar size
Q4_K_M · 67.7 GB ~4.4 CPU offload
gpt-oss 120B
117B MoE · similar size
MXFP4 · 61.4 GB ~6.1 CPU offload
Llama 4 Scout 109B-A17B
109B MoE · similar size
Q4_K_M · 63.4 GB ~1.7 CPU offload
Qwen3.8-Flash-Next 180B-A6B
180B MoE · similar size
Q4_K_M · 102.0 GB ~4.5 CPU offload

Every model against the Arc B580 →

Ling 3.0 Flash 124B-A5B on other hardware

DeviceTotalTok/s est.
GeForce RTX 4060 Laptop
7.0 GB usable · 256 GB/s
70.4 GB ~5.2 CPU offload
GeForce RTX 3060 12 GB
10.6 GB usable · 360 GB/s
70.4 GB ~5.5 CPU offload
M4 · 16 GB
10.7 GB usable · 120 GB/s
70.4 GB — Won't fit
GeForce RTX 4090
22.4 GB usable · 1008 GB/s
70.4 GB ~6.9 CPU offload
Radeon RX 7900 XTX
22.4 GB usable · 960 GB/s
70.4 GB ~6.9 CPU offload
M3 Max · 64 GB
48.0 GB usable · 400 GB/s
70.4 GB — Won't fit
DGX Spark · 128 GB
96.0 GB usable · 273 GB/s
70.4 GB 25 Runs great
H200 SXM
139.4 GB usable · 4800 GB/s
70.4 GB 390 Runs great

All 114 devices, by tier →