What hardware do you need to run Ling 3.1 flash?

Ling 3.1 flash at Q4_K_M is 358 GB of weights, with a context ceiling of 256k tokens. 25B active out of 560B, built for long agent runs, search and software engineering. Only nine of its 72 layers cache the whole context, so almost all of the memory it asks for goes on weights. inclusionAI published a post-training update, Ling-3.1-flash-1019, on 27 October; it is the same size and architecture, and the four-bit build priced here is of the original.

How good is Ling 3.1 flash, really?

On the Artificial Analysis Intelligence Index v4.3 it scores 41 (verbose: 220M output tokens over the index against a median of 140M), which puts it in the Sonnet-class band. In the same league as the labs' mainstream models on this index. See the whole table.

Head to head with its neighbours on the leaderboard: vs GLM-5.3-Flash above it, vs Qwen3.8 Flash Next below.

What Ling 3.1 flash costs either way

Renting the same model costs $0.075 per million input tokens and $0.225 per million output (OpenRouter, cheapest of two active endpoints: DeepInfra and Novita tied on it, checked 2026-10-27). Buying a machine only beats that if you use it hard enough, for long enough, that the hardware price divides down below the rental bill. What a million tokens costs each way puts the two prices side by side.

The machines that miss Ling 3.1 flash, and what they run

None of the machines here with a published price that are still sold holds Ling 3.1 flash at any window, and none of them misses by a window: its 358 GB of weights are larger than the usable memory in every one of them, before a single token of key-value cache. The Mac Studio M5 Ultra, 256GB comes nearest, 166 GB short at 4k, the shortest window on the list. Qwen3.8 27B is the strongest model six of the eight hold, and it scores 34 on the intelligence index against this model's 41.

Two machines here do hold Ling 3.1 flash at 32k, and neither of them is an answer to what to buy, which is the question the rest of this page asks: the Mac Studio M3 Ultra, 512GB, no longer sold, at the $9,499 it launched at, and the Mac Studio M5 Ultra, 512GB, on sale with no published price. They are sold with 512 GB of memory and hand a model 384 GB of it, against the 358 GB Ling 3.1 flash needs at 32k.

One machine per family, the largest configuration in each that still misses, nearest first. Short by is the weights plus the key-value cache at 4k measured against the memory that machine's GPU can address, which that machine's own page gives. The last column is the highest-scoring current model the machine holds at 32k.

Price the Mac Studio M5 Ultra, 256GB on GLM-5.3-Flash instead

The specifics

Parameters

560B, of which 25B are active per token

Quantisation

Q4_K_M

Weights on disk

358 GB

KV cache

0.3 GB at 32k context — 63 of the 72 layers are Kimi Delta Attention, a linear-attention layer whose state is a fixed size and does not grow with the context. The other 9 use gated latent attention: each caches one 512-number compressed vector plus a 64-number rotary key per token, 1,152 bytes a layer at 16 bits, so 10,368 bytes a token for the whole model. inclusionAI publishes the weights in FP8, so the four-bit GGUF is a requant of an already-quantised release. How weights and cache add up.

Maximum context

256k tokens (the card says 1M)

Licence

MIT

Sources

inclusionAI/Ling-3.1-flash, bartowski/Ling-3.1-flash-GGUF