Inclusion AILingBailingMoeV3ForCausalLM

Ling 3.0 Flash Fin — Hardware Requirements & GPU Compatibility

ChatFunctions

Ling 3.0 Flash Fin is a 127.5B-parameter open language model from Inclusion AI in the Ling family. It supports a context window of up to 262,144 tokens. At Q4_K_M it needs about 77.67 GB of VRAM — see which GPUs and Macs can run it below.

127.5B params · MoE5.6B active262K contextMITSep 20261.8K downloads107 likes

Based on Ling 3.0 Flash

Your hardware

Verdicts below are for the reference RTX 4090. Set yours to personalise the page.

Detection reads your browser's GPU name with no data sent anywhere.

Ling 3.0 Flash Fin runs on a 96 GB+ Mac or a workstation GPU

Best local pick → Mac Studio (M3 Ultra, 96GB) · Q4_K_M · 77.6 GB · ~70 tok/s

Macs from 64 GB (Q2_K · ~79 tok/s)

MIT

Not set up for it?

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

Run Ling 3.0 Flash Fin

One recommended way. Every other method, OS and quant is in the step-by-step guide.

Other ways

each opens step-by-step

Run in cloud

Fits on RTX PRO 6000 (96 GB) (18 GB headroom) · Q4_K_M

Generation speed: ~179 tok/s; generation speed
Cost per 1M output tokens: $2.39; per 1M output tokens

Deployon Vast.aiBest optionfrom $1.54/hr

Compare GPUs →

Run Ling 3.0 Flash Fin

77.6 GB VRAM at 16K context · needs offload on the RTX 4090 (53.6 GB short)

Ollama is not available

No GGUF repository is listed for this model.

Linux only.

Install vLLM
Linux only; on Windows run it inside WSL2. NVIDIA GPU shown; AMD ROCm, Intel and CPU builds have their own install pages.
$uv venv --python 3.12 --seed
$source .venv/bin/activate
$uv pip install vllm --torch-backend=auto
vLLM installation ↗
Serve the model
OpenAI-compatible API at http://localhost:8000/v1.
$vllm serve inclusionAI/Ling-3.0-flash-Fin --max-model-len 16384
Runs the original weights, not the Q4_K_M GGUF; needs ~280 GB of GPU memory.

Rent GPUs hourly to run Q4_K_M right there. Pricing comes from partner marketplaces and varies.

Available GPU rentals for Q4_K_M, prioritized by value
GPU	Speed	$/1M tokens	Rent
RTX PRO 6000 Server (96 GB)CheapestCheapest18 GB free	~179 tok/s	$2.39	Deploy on Vast.aifrom $1.54/hr (opens in new tab)	96 GB · 18 GB free · ~179 tok/s · $2.39 / 1M tokens
RTX PRO 6000 Max-Q (96 GB)18 GB free	~191 tok/s	$2.28	Deploy on Vast.aifrom $1.56/hr (opens in new tab)	96 GB · 18 GB free · ~191 tok/s · $2.28 / 1M tokens
RTX PRO 6000 Server (96 GB)18 GB free	~179 tok/s	$2.63	Deploy on RunPodfrom $1.69/hr (opens in new tab)	96 GB · 18 GB free · ~179 tok/s · $2.63 / 1M tokens
RTX PRO 6000 (96 GB)18 GB free	~191 tok/s	$2.72	Deploy on Vast.aifrom $1.87/hr (opens in new tab)	96 GB · 18 GB free · ~191 tok/s · $2.72 / 1M tokens
H200 SXM (141 GB)63 GB free	~294 tok/s	$3.40	Deploy on RunPodfrom $3.59/hr (opens in new tab)	141 GB · 63 GB free · ~294 tok/s · $3.40 / 1M tokens
H200 NVL (141 GB)63 GB free	~294 tok/s	$3.92	Deploy on Vast.aifrom $4.15/hr (opens in new tab)	141 GB · 63 GB free · ~294 tok/s · $3.92 / 1M tokens
B200 (192 GB)114 GB free	~337 tok/s	$4.93	Deploy on RunPodfrom $5.98/hr (opens in new tab)	192 GB · 114 GB free · ~337 tok/s · $4.93 / 1M tokens
B300 (288 GB)210 GB free	~337 tok/s	$5.72	Deploy on RunPodfrom $6.94/hr (opens in new tab)	288 GB · 210 GB free · ~337 tok/s · $5.72 / 1M tokens

Compare GPUs →

How Much VRAM Does Ling 3.0 Flash Fin Need?

Select a quantization level; the listed hardware adheres to it.

VRAM	+32K	+128K	File	Quality	On RTX 4090 (reference)
55.4 GB	56.1 GB	60.8 GB	~54.2 GB	2-bit · K-quant — 2-bit quantization with K-quant improvements	Does not fit: needs 55.3 GB
63.3 GB	64 GB	68.8 GB	~62.2 GB	3-bit medium — 3-bit medium quantization	Does not fit: needs 63.2 GB
77.7 GB	78.4 GB	83.1 GB	~76.5 GB	4-bit medium — 4-bit medium quantization — most popular sweet spot	Does not fit: needs 77.6 GB
92.0 GB	92.7 GB	97.5 GB	~90.8 GB	5-bit medium — 5-bit medium quantization — good quality/size tradeoff	Does not fit: needs 91.9 GB
106.4 GB	107.1 GB	111.8 GB	~105.2 GB	6-bit, very good — 6-bit quantization, very good quality	Does not fit: needs 106.3 GB
128.7 GB	129.4 GB	134.1 GB	~127.5 GB	8-bit, near-lossless — 8-bit quantization, near-lossless	Does not fit: needs 128.6 GB
256.1 GB	256.9 GB	261.6 GB	~255 GB	16-bit · bfloat — Brain floating point 16 — preferred for training	Does not fit: needs 256.1 GB

est.calculated, no published file yet

Selected Q4_K_M

4-bit medium quantization

77.6GBat 16K context

Weights: 76.5 GB
KV cache: 0.8 GB
Overhead: 0.3 GB

Doesn't fit RTX 4090 (reference) · needs 53.6 GB more

≥50 tok/s needs ≥297 GB/s · M3 Ultra ~70, M2 Ultra ~69

Or rent RTX PRO 6000 from $1.54/hr.

Which Hardware Can Run Ling 3.0 Flash Fin?

Q4_K_M · 77.7 GB

Ling 3.0 Flash Fin (Q4_K_M) requires 77.7 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 101+ GB is recommended. Using the full 262K context window can add up to 11.8 GB, bringing total usage to 89.5 GB. No consumer GPU has enough memory. 30 devices with unified memory can run Ling 3.0 Flash Fin, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Studio (M3 Ultra, 96GB).

By VRAM tier · typical GPUs and Macs at Q4_K_M

Typical hardwareQuantVRAMtok/sContext that fits

8–16 GB · RTX 4060, RTX 3060 12GB, RTX 5060 Ti 16GBDoesn't fit — smallest quant needs 55.3 GB8 GB tier →12 GB tier →16 GB tier →

24 GB · RTX 4090referenceDoesn't fit — smallest quant needs 55.3 GB24 GB tier →

Mac 32 GB · Mac Mini M4Doesn't fit — smallest quant needs 55.3 GBMacs →

Mac 64 GB · Mac Studio M4 MaxQ2_K · 55.3 GB · ~68 tok/s · 64K55.3~6864KMacs →

Alternatives to Ling 3.0 Flash Fin

Ling family · 4 models →

Similar models

4tok/s on RTX 4090

Model	Score	VRAM	tok/s	Why it is similar
Qwen3.5 122B A10BAlibaba	109.4	75.5 GB	—needs > 24 GB	same size
NVIDIA Nemotron 3.5 Super VL 120B A12B BF16NVIDIA	—	87.7 GB	—needs > 24 GB	same size
Leanstral 1.5 119B A6BMistral AI	—	84 GB	—needs > 24 GB	same size
XORTRON.CriminalComputing.LARGE.2026.3darkc0de	—	79.8 GB	—needs > 24 GB	same size

Specifications

Publisher: Inclusion AI
Family: Ling
Parameters: 127.5B
Active parameters: 5.6B per token
Architecture: BailingMoeV3ForCausalLM
Context Length: 262,144 tokens
Vocabulary Size: 157,184
Release Date: 2026-09-03
License: MIT

Resources

Frequently Asked Questions

How much VRAM does Ling 3.0 Flash Fin need?

Ling 3.0 Flash Fin requires 77.7 GB of VRAM at Q4_K_M, or 256.1 GB at BF16. Full 262K context adds up to 11.8 GB (89.5 GB total).

VRAM = Weights + KV Cache + Overhead

Weights = 127.5B × 4.8 bits ÷ 8 = 76.5 GB

KV Cache + Overhead ≈ 1.2 GB (at 2K context + ~0.3 GB framework)

Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.

KV Cache + Overhead ≈ 13 GB (at full 262K context)

VRAM usage by quantization

Q4_K_M

77.7 GB

Q4_K_M + full context

89.5 GB

Learn more about VRAM estimation →

Can NVIDIA GeForce RTX 5090 run Ling 3.0 Flash Fin?

No — Ling 3.0 Flash Fin requires at least 55.4 GB at Q2_K, which exceeds the NVIDIA GeForce RTX 5090's 32 GB of VRAM.

What's the best quantization for Ling 3.0 Flash Fin?

For Ling 3.0 Flash Fin, Q4_K_M (77.7 GB) offers the best balance of quality and VRAM usage. Q5_K_M (92.0 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 55.4 GB.

VRAM requirement by quantization

Q2_K

55.4 GB~75%

Q4_K_M ★

77.7 GB~89%

Q5_K_M

92.0 GB~92%

Q6_K

106.4 GB~95%

Q8_0

128.7 GB~99%

BF16

256.1 GB~100%

★ Recommended — best balance of quality and VRAM usage.

Learn more about quantization →

Can I run Ling 3.0 Flash Fin on a Mac?

Yes — Mac Pro M2 Ultra (64 GB) and 41 other Macs can run Ling 3.0 Flash Fin, the smallest at Q2_K. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 77.7 GB of usable unified memory (RAM minus macOS overhead), from Mac Studio (M3 Ultra, 96GB).

Can I run Ling 3.0 Flash Fin locally?

Yes — Ling 3.0 Flash Fin can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is Ling 3.0 Flash Fin?

At Q4_K_M, Ling 3.0 Flash Fin can reach ~110 tok/s on AMD Instinct MI350X. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.

tok/s = 1000 ÷ (active GB ÷ (bandwidth GB/s × efficiency) × 1000 + layers × routing ms)

Mixture-of-Experts: just the active experts get read for each token, along with a fixed routing cost per layer.

Example: NVIDIA B200 → 3.4 GB active ÷ (8000 × 0.65) = 0.66 ms, plus 42 layers × 0.055 ms = 2.31 ms, so 1000 ÷ 2.97 ms = ~337 tok/s

Estimated speed at Q4_K_M (77.7 GB)

NVIDIA B200

~337 tok/s

NVIDIA B300

~337 tok/s

NVIDIA GH200 Grace Hopper Superchip

~296 tok/s

Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.

Learn more about tok/s estimation →

What's the download size of Ling 3.0 Flash Fin?

At Q4_K_M, the download is about 76.49 GB. The full-precision BF16 version is 254.97 GB. The smallest option (Q2_K) is 54.18 GB.

Which GPUs can run Ling 3.0 Flash Fin?

No single consumer GPU has enough VRAM to run Ling 3.0 Flash Fin at Q4_K_M (77.7 GB). Multi-GPU or professional hardware is required.

Which devices can run Ling 3.0 Flash Fin?

30 devices with unified memory can run Ling 3.0 Flash Fin at Q4_K_M (77.7 GB), including ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB), GMKtec EVO-X2 (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.

Ling 3.0 Flash Fin — Hardware Requirements & GPU Compatibility

Ling 3.0 Flash Fin runs on a 96 GB+ Mac or a workstation GPU

Run Ling 3.0 Flash Fin

Recommended

Other ways

Run in cloud

Run Ling 3.0 Flash Fin

Install vLLM

Serve the model

All sources

How Much VRAM Does Ling 3.0 Flash Fin Need?

Selected Q4_K_M

Which Hardware Can Run Ling 3.0 Flash Fin?

Alternatives to Ling 3.0 Flash Fin

Similar models

Specifications

Resources

Frequently Asked Questions

Run Ling 3.0 Flash Fin

Install vLLM

Serve the model

All sources