Llama 4 Maverick TPS calculator

Open weights Meta AI 400B parameters April 2025

Each card below is assessed against this model at the context length and minimum quality you choose. Speed is an estimate for a single request, calculated from the card's memory bandwidth and the size of the model once compressed.

Calculated for this model

6 of 818 cards that can run it

Smallest card that fits

Radeon Instinct MI300X

192 GB · Q3_K_M · 65.9 tok/s

Fastest card

B300

109 tok/s · 288 GB

Which GPUs can run Llama 4 Maverick?

Set the inputs, read the answer

A longer conversation needs more memory, which can push this model off smaller cards.

Hides cards that would only fit the model by compressing it below this point.

6 cards match

Calculating
Needs Quantisation Fit
109 tok/s

65–174 · low confidence

B300 NVIDIA 288 GB 8,000 GB/s Sep 2025 214.8 GB Q4_K_M Comfortable
86.8 tok/s

52–139 · low confidence

Radeon Instinct MI350X AMD 288 GB 8,190 GB/s Jan 2025 214.8 GB Q4_K_M Comfortable
86.8 tok/s

52–139 · low confidence

Radeon Instinct MI355X AMD 288 GB 8,190 GB/s Jan 2025 214.8 GB Q4_K_M Comfortable
65.9 tok/s

40–105 · low confidence

Radeon Instinct MI300X AMD 192 GB 5,325 GB/s Dec 2023 168.2 GB Q3_K_M Tight
65.9 tok/s

40–105 · low confidence

Radeon Instinct MI308X AMD 192 GB 5,325 GB/s Dec 2023 168.2 GB Q3_K_M Tight
63.6 tok/s

38–102 · low confidence

Radeon Instinct MI325X AMD 256 GB 6,000 GB/s Oct 2024 214.8 GB Q4_K_M Tight

Speeds are estimates for a single request — one conversation at a time — calculated from memory bandwidth, model size and quantisation. Real throughput varies with the inference runtime and its version. Figures published by hardware vendors measure many simultaneous requests and are much higher.

On record

Full specification

Everything on record for this model. Most of it describes how it was trained rather than how it runs — useful context for judging how much work went into it, and how it compares with models built at a different scale.

Origin

Who built this model, where, and when it was published.

Organisation
Meta AI
Organisation type
Industry
Country
United States of America
Published
5 April 2025

What it does

The problem areas the model was built for. A model can carry several of each.

Domain
Multimodal, Language, Vision
Task
Chat, Code generation, Visual question answering, Language modeling/generation, Question answering

Size

How large the model is and how much data it was trained on. Parameters are the figure that decides whether it fits on a given graphics card.

Parameters
400B

"Llama 4 Maverick models have 17B active parameters and 400B total parameters." https://ai.meta.com/blog/llama-4-multimodal-intelligence/

Training data
30,000,000,000,000 tokens

"The overall data mixture for training consisted of more than 30 trillion tokens, which is more than double the Llama 3 pre-training mixture and includes diverse text, image, and video datasets."

Training compute

The arithmetic performed to train the model, measured in floating-point operations. It is a measure of what the training run cost, not of how fast the finished model answers you.

Training compute
2.2 × 10²⁴ FLOP

22T training tokens per model card: https://github.com/meta-llama/llama-models/blob/main/models/llama4/MODEL_CARD.md Maverick was trained using co-distillation from Llama 4 Behemoth. It isn't 100% clear that all 22T tokens used distillation, but we assume this for the time being. Estimating training compute from parameters and tokens: Compute = 6 FLOP per token per parameter * 17B active parameters * 22T tokens = 2.244e24 FLOP (Implying mean throughput was 262 TFLOPS/GPU, or 13.2% MFU in F…

Availability

Whether you can obtain the model and run it on your own hardware, which is what decides if any of the graphics-card figures on this page apply.

Weights
Open — downloadable
Model access
Open weights (restricted use)
Training code
Unreleased

Llama 4 license (branding requirements, size cap 700M MAU) https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Original no training code here https://github.com/meta-llama/llama-models/tree/main/models/llama4

Hugging Face
meta-llama

How it is classified

Labels the source dataset applies when tracking notable models, and how confident it is in the entry.

Likely above 10²³ FLOP
Yes
Why it is tracked
Discretionary
Record confidence
Likely

Sources

Where this record came from and when it was last checked.

Reference
The Llama 4 herd: The beginning of a new era of natively multimodal AI innovation
Last updated
28 November 2025

What the numbers mean

Hardware requirements in practice

Minimum card

Radeon Instinct MI300X

Memory needed

168.2 GB

Fastest

109 tok/s

At 400B parameters, Llama 4 Maverick is beyond what any single graphics card holds. Running it means either splitting it across several cards or renting hardware built for the job — 6 of the cards we track can hold it on their own, and all of them are datacentre parts.

The entry point is the Radeon Instinct MI300X: 192 GB of memory, Q3_K_M compression, roughly 65.9 tokens per second.

A B300 is the fastest we calculate for it: about 109 tokens per second, from 8,000 GB/s of memory bandwidth.

What this model is

Llama 4 Maverick was published by Meta AI, in United States of America, in April 2025. The organisation is categorised as industry.

It works in Multimodal, Language, Vision, and is recorded as doing chat, Code generation, Visual question answering, Language modeling/generation, Question answering.

The weights are published, so it can be downloaded and run on your own hardware indefinitely, offline, with no account attached. It is published under the meta-llama organisation on Hugging Face.

What decides the speed

The median result is around 76.3 tokens per second; 6 cards produce text faster than most people read it.

Mixture-of-experts routing is why the speeds here look high for the parameter count. Only a fraction is read per token, but the whole thing has to be loaded.

Memory here is estimated from size rather than computed from the architecture, which is not recorded for this model — the numbers are indicative rather than exact.

Training and provenance

Producing it required around 2.2 × 10²⁴ FLOP of arithmetic, which is a statement about the training budget rather than about inference.

It was trained on about 30,000,000,000,000 tokens of text.

Its inclusion criterion is discretionary.

Step by step

How to choose a GPU for Llama 4 Maverick

The table above has already assessed every card we hold specifications for against this model. Getting to your answer takes six steps.

  1. 01

    Read the memory figure first

    The table lists every card that can hold Llama 4 Maverick — around 168.2 GB at Q3_K_M. That figure, not the card's headline performance, is what decides whether it runs.

  2. 02

    Match the context to your actual use

    Set the context to what you will actually use. The cache grows with the conversation, and it is the usual reason Llama 4 Maverick stops fitting a card that seemed fine.

  3. 03

    Decide how much compression you will accept

    The quantisation column varies by card, because a bigger card holds a more accurate copy of Llama 4 Maverick — Q3_K_M on the smallest card that fits. Set a floor to hold the comparison at one level.

  4. 04

    Rank by throughput rather than spec sheet

    Ranking by tokens per second for Llama 4 Maverick follows memory bandwidth, not core counts, which is why the B300 tops it at 109 tok/s.

  5. 05

    Read the fit column last

    A tight fit runs Llama 4 Maverick but leaves nothing spare for a longer conversation; comfortable has headroom. If you expect to grow the context, buy for comfortable.

  6. 06

    See what else that card runs

    Each card page repeats this sweep for every model we hold. It answers what else the hardware is good for, beyond Llama 4 Maverick.

Answers

Llama 4 Maverick — common questions

01

How much compute was used to train Llama 4 Maverick?

Around 2.2 × 10²⁴ FLOP. 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.

02

Can I run Llama 4 Maverick if it does not fit in my GPU?

Partly. Layers that do not fit sit in system memory and run at a fraction of the speed, so a mostly-offloaded Llama 4 Maverick is rarely worth using — the nearest miss we calculate is short by 52.8 GB. Every figure here assumes the whole model is on the card.

03

Would two GPUs run Llama 4 Maverick faster?

Capacity adds across cards; throughput does not. Since 6 of the cards we track already hold Llama 4 Maverick on their own, a second card is rarely the answer here.

04

Why does the quantisation differ between cards for Llama 4 Maverick?

A larger card holds a more accurate copy. Across the cards that run Llama 4 Maverick, 2 compression levels are used; the floor control above pins it to one.

05

How accurate are these Llama 4 Maverick speed estimates?

They are calculated from specifications rather than measured, and each carries a range — 65–174 tok/s on the B300, for instance. The same model and card vary by thirty to fifty per cent depending on the inference software and its version.

06

What GPU do I need to run Llama 4 Maverick?

The smallest card in our catalogue that holds Llama 4 Maverick is the Radeon Instinct MI300X, with 192 GB of memory. It runs the model at Q3_K_M using about 168.2 GB, and produces roughly 65.9 tokens per second. 6 cards in total can run it.

07

How fast is Llama 4 Maverick on a GPU?

It depends on the card. The quickest we calculate is a B300 at about 109 tokens per second; the slowest that still runs it manages considerably less. Reading speed is around ten tokens per second, and 6 of the cards that can run Llama 4 Maverick clear that.

08

How much VRAM does Llama 4 Maverick need?

About 168.2 GB at Q3_K_M compression, 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.

09

Is Llama 4 Maverick open source?

Its weights are published, so Llama 4 Maverick 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.

10

How many parameters does Llama 4 Maverick have?

Llama 4 Maverick has 400B parameters. "Llama 4 Maverick models have 17B active parameters and 400B total parameters." https://ai.meta.com/blog/llama-4-multimodal-intelligence/. 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.

11

Who created Llama 4 Maverick?

Llama 4 Maverick was published by Meta AI, based in United States of America, categorised as industry.

12

When was Llama 4 Maverick released?

Llama 4 Maverick was published in April 2025.

13

What is Llama 4 Maverick used for?

Llama 4 Maverick works in Multimodal, Language, Vision, and is recorded as handling chat, Code generation, Visual question answering, Language modeling/generation, Question answering. A model can carry several of each, so these are the areas it was built for rather than a limit on what it will attempt.

14

Where can I download Llama 4 Maverick?

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

Source

Original publication

Record last updated 28 November 2025

The other direction

Looking at it from the other side?

This page starts from the model. If you already own a card and want to know everything it will run, start from the hardware instead.