What hardware do you need to run Ripple 65B Preview?

Ripple 65B Preview at Q4_K_M is 39 GB of weights, with a context ceiling of 1M tokens. Every parameter runs on every token: this is a dense model, where most of what the open labs ship at this strength is a mixture of experts. Ricursive built it thin and deep, 80 narrow layers rather than fewer wide ones, and put four layers on a short 1,024-token window for every layer that caches the whole context. That ratio is the reason a model this size can hold a long window on a desktop machine, and Ricursive says it fits an NVIDIA DGX Spark, which the machines listed here bear out. Text only, trained mostly on English and code. The speeds here are worked out from memory bandwidth, the way every estimate on this site is; the one figure Ricursive publishes is 250 tokens a second on two B200s at batch size 8, which is neither a machine priced here nor one request at a time. On the cheapest machine that runs it, a Corsair AI Workstation 300, 64GB, it generates 4.7 tok/s at 32k of context, worked out from memory bandwidth.

How good is Ripple 65B Preview, really?

It has not been placed on the intelligence index yet. See the whole table.

What Ripple 65B Preview costs either way

Renting the same model costs $0.2 per million input tokens and $0.6 per million output (OpenRouter, the only active endpoint, checked 2026-11-06). 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.

Machines that run Ripple 65B Preview

Every machine here runs it, but not to the same length. The Strix Halo Corsair AI Workstation 300, 64GB stops at 128k; the DGX Spark GB10 Grace Blackwell, 128GB takes it to 256k, the longest the calculator offers. The weights are the same size on every machine; what differs is the memory left for the key-value cache, which grows with every token you keep.

The table is the cheapest machine in each family, not the only one that runs it: 28 of the 56 machines on this site hold it at 32k. Within a family only memory decides, so that is every Mac Studio, Strix Halo box, Mac mini and MacBook Pro with 64GB or more, every NVIDIA card with 96GB or more and the DGX Spark, 128GB. One table sets every machine against the others: price, the memory its GPU can use, what it holds and how long it takes to pay for itself.

One machine per family, cheapest first. Speeds are measured where a public benchmark exists and estimated from memory bandwidth otherwise; the calculator says which for any configuration. The longest context is the longest setting the calculator offers that the machine still holds this model at, cache included, and no machine is shown taking it past its own 1M limit. Each one opens the calculator on that machine at that length.

For scale, the calculator starts from 80 tok/s for a hosted API and times a local machine against it. Nothing above reaches it, and the quickest is 35 tok/s.

The cheapest machine that runs it is sold with 64 GB. There is a page on what 64 GB runs, machine by machine: it names the 9 machines here sold at that size, and the 37 current models the roomiest of them holds at 32k, this one among them.

Run Ripple 65B Preview at 256k on the DGX Spark GB10 Grace Blackwell, 128GB

How much memory does Ripple 65B Preview need at 256k?

Every figure above is priced at 32k of context. At 256k, the longest the calculator offers, Ripple 65B Preview needs 56.2 GB rather than 41.2 GB. The weights are the same either way; the key-value cache goes from 2.4 GB to 17.4 GB.

20 of the 37 machines here still sold at a published price hold Ripple 65B Preview at 32k and 13 hold it at 256k. The cheapest still sold goes from the Corsair AI Workstation 300, 64GB at $1,700 to the Framework Desktop, 128GB at $3,449.

On the Framework Desktop, 128GB it runs 4.7 tok/s estimated at 32k and 3.4 tok/s estimated at 256k. Every token generated reads the whole cache back, so a longer window is slower as well as bigger.

The specifics

Parameters

64.8B

Quantisation

Q4_K_M

Weights on disk

39 GB

KV cache

2.4 GB at 32k context — Four layers on a 1,024-token window to every one that caches the whole context, so 16 of the 80 grow and the other 64 stop. All 80 carry eight KV heads at 128 wide, the same as Llama 3.3 70B, which caches every token in all 80 of its own layers: the window is the whole of the difference between them. How weights and cache add up.

Maximum context

1M tokens (long-context training ran at twice this, and the publisher supports the shorter figure)

Licence

Apache 2.0

Sources

ricursiveAI/Ripple-65B-Preview, ricursiveAI/Ripple-65B-Preview (config), AtomicChat/Ripple-65B-Preview-GGUF, mrfakename/Ripple-65B-Preview-GGUF, openrouter.ai