Motif-3 TPS calculator

It needs about 154.1 GB at a compression of Q3_K_M, which is what the smallest card that runs it uses. Less compression needs more: the figures in the memory column above are recalculated for each card, because each one holds the least-compressed version it can.

Its weights are published, so it can be downloaded and run on your own hardware. Note that open weights is not the same as open source in the full sense — it says nothing about the training data, the training code, or the commercial terms attached.

It has a parameter count of 314B. Fine-grained sparse MoE: approximately 314B total parameters, ~13.2B activated per token. 53 Transformer layers (2 dense + 51 MoE); each MoE layer has 384 routed experts + 1 shared expert, with top-8 routing. Grouped Differential Latent Attention (GDLA), modified manifold-constrained hyper-connections, Expert-Specific PolyNorm, and a 1-layer MTP head. Trained from scratch. That figure is the total, and it is what decides how much memory the model needs — roughly half a gigabyte per billion at the compression most people use.

It was published by Motif Technologies, based in Korea (Republic of), an organisation categorised as industry.

It was published in August 2026.

It works in the domain of Language, and is recorded as handling the task of language modeling/generation, Chat, Question answering, Code generation, Mathematical reasoning, Instruction interpretation. Models frequently carry more than one of each, and the tags describe purpose rather than capability limits.

Its weights are published on Hugging Face, under the organisation Motif-Technologies. We do not host model files — this site calculates what hardware is needed to run them.

Training consumed around 1.1 × 10²⁴ FLOP, on hardware recorded as NVIDIA B200. That measures what producing the model cost and says nothing about how quickly it answers once trained — inference speed comes from memory bandwidth, not from the training budget.

Partly. Layers that do not fit sit in system memory and run at a fraction of the speed, so a mostly-offloaded model is rarely worth using. The nearest miss we calculate falls short by 63.8 GB. Every figure here assumes the whole model is resident on the card.

A second card roughly doubles the memory available but not the generation rate. The number already able to run it alone: 7. So a second card is rarely the answer here.

A larger card holds a more accurate copy. The number of compression levels used across the cards that run it: 3. Bigger cards get the more accurate version, and the quality floor above pins the comparison to one level.

Every figure is derived from memory bandwidth and model size, not benchmarked, which is why each is published as a range rather than a single number. One example: 17–47 tok/s on B200. The same model and card vary by thirty to fifty per cent depending on the inference software and its version.

The smallest card in our catalogue that holds it is B200, with a memory capacity of 180 GB. It runs the model at a compression of Q3_K_M using about 154.1 GB, and produces roughly 29.1 tokens per second. The number of cards able to run it in total: 7.

It depends on the card. The quickest we calculate is B200, at about 29.1 tokens per second; the slowest that still runs it manages considerably less. Reading speed is around ten tokens per second, and the number of cards clearing that: 7.