Kosmos-2.5 TPS calculator

Open weights Microsoft 1.3B parameters August 2024

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

818 of 818 cards that can run it

Smallest card that fits

Tesla C1080

4 GB · Q8_0 · 28.4 tok/s

Fastest card

B200

2,606 tok/s · 180 GB

Which GPUs can run Kosmos-2.5?

Set the inputs, read the answer

Context length tokens

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

Minimum quality

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

Search cards

Card memory

Also show cards that cannot run it

818 cards match

Calculating

					Needs	Quantisation	Fit
2,606 tok/s 1,564–4,170 · low confidence	B200 NVIDIA	180 GB	8,000 GB/s	Jan 2024	2.1 GB	Q8_0	Comfortable
2,606 tok/s 1,564–4,170 · low confidence	B300 NVIDIA	288 GB	8,000 GB/s	Sep 2025	2.1 GB	Q8_0	Comfortable
2,081 tok/s 1,249–3,330 · low confidence	Radeon Instinct MI350X AMD	288 GB	8,190 GB/s	Jan 2025	2.1 GB	Q8_0	Comfortable
2,081 tok/s 1,249–3,330 · low confidence	Radeon Instinct MI355X AMD	288 GB	8,190 GB/s	Jan 2025	2.1 GB	Q8_0	Comfortable
1,664 tok/s 999–2,663 · low confidence	Radeon Instinct MI300 AMD	128 GB	6,550 GB/s	Jan 2023	2.1 GB	Q8_0	Comfortable
1,593 tok/s 956–2,549 · low confidence	H200 NVL NVIDIA	141 GB	4,890 GB/s	Nov 2024	2.1 GB	Q8_0	Comfortable
1,593 tok/s 956–2,549 · low confidence	H200 SXM 141 GB NVIDIA	141 GB	4,890 GB/s	Nov 2024	2.1 GB	Q8_0	Comfortable
1,525 tok/s 915–2,440 · low confidence	Radeon Instinct MI325X AMD	256 GB	6,000 GB/s	Oct 2024	2.1 GB	Q8_0	Comfortable
1,353 tok/s 812–2,165 · low confidence	Radeon Instinct MI300A AMD	128 GB	5,325 GB/s	Dec 2023	2.1 GB	Q8_0	Comfortable
1,353 tok/s 812–2,165 · low confidence	Radeon Instinct MI300X AMD	192 GB	5,325 GB/s	Dec 2023	2.1 GB	Q8_0	Comfortable
1,353 tok/s 812–2,165 · low confidence	Radeon Instinct MI308X AMD	192 GB	5,325 GB/s	Dec 2023	2.1 GB	Q8_0	Comfortable
1,284 tok/s 770–2,054 · low confidence	H100 NVL 94 GB NVIDIA	94 GB	3,940 GB/s	Mar 2023	2.1 GB	Q8_0	Comfortable
1,095 tok/s 657–1,751 · low confidence	H100 PCIe 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	2.1 GB	Q8_0	Comfortable
1,095 tok/s 657–1,751 · low confidence	H100 SXM5 80 GB NVIDIA	80 GB	3,360 GB/s	Oct 2022	2.1 GB	Q8_0	Comfortable
1,095 tok/s 657–1,751 · low confidence	H100 SXM5 94 GB NVIDIA	94 GB	3,360 GB/s	Mar 2023	2.1 GB	Q8_0	Comfortable
1,095 tok/s 657–1,751 · low confidence	H100 SXM5 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	2.1 GB	Q8_0	Comfortable
1,095 tok/s 657–1,751 · low confidence	H800 SXM5 NVIDIA	80 GB	3,360 GB/s	Mar 2023	2.1 GB	Q8_0	Comfortable
834 tok/s 500–1,334 · low confidence	Radeon Instinct MI250 AMD	128 GB	3,280 GB/s	Nov 2021	2.1 GB	Q8_0	Comfortable
834 tok/s 500–1,334 · low confidence	Radeon Instinct MI250X AMD	128 GB	3,280 GB/s	Nov 2021	2.1 GB	Q8_0	Comfortable
695 tok/s 417–1,111 · low confidence	Data Center GPU Max 1550 Intel	128 GB	3,280 GB/s	Jan 2023	2.1 GB	Q8_0	Comfortable
680 tok/s 408–1,088 · low confidence	Data Center GPU Max Subsystem Intel	128 GB	3,210 GB/s	Jan 2023	2.1 GB	Q8_0	Comfortable
665 tok/s 399–1,063 · low confidence	A100 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Nov 2020	2.1 GB	Q8_0	Comfortable
665 tok/s 399–1,063 · low confidence	A100X NVIDIA	80 GB	2,040 GB/s	Jun 2021	2.1 GB	Q8_0	Comfortable
665 tok/s 399–1,063 · low confidence	A800 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Aug 2022	2.1 GB	Q8_0	Comfortable
665 tok/s 399–1,063 · low confidence	H100 CNX NVIDIA	80 GB	2,040 GB/s	Mar 2023	2.1 GB	Q8_0	Comfortable

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: Microsoft
Organisation type: Industry
Country: United States of America
Published: 21 August 2024
Authors: Tengchao Lv, Yupan Huang, Jingye Chen, Yuzhong Zhao, Yilin Jia, Lei Cui, Shuming Ma, Yaoyao Chang, Shaohan Huang, Wenhui Wang, Li Dong, Weiyao Luo, Shaoxiang Wu, Guoxin Wang, Cha Zhang, Furu Wei

What it does

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

Domain: Multimodal, Language, Vision
Task: Character recognition (OCR), Document classification, Language modeling/generation, Visual question answering, Document representation
Base model: Pix2Struct-Large

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: 1.3B
Training data: 260,000,000,000 tokens

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 × 10²¹ FLOP
How it was established: Operation counting

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 (unrestricted)
Training code: Unreleased
Hugging Face: microsoft

How it is classified

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

Record confidence: Confident

Sources

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

Reference: KOSMOS-2.5: A Multimodal Literate Model
Last updated: 28 November 2025

The extremes

The ten fastest GPUs for Kosmos-2.5

Ranked by estimated tokens per second, newest card first where speeds tie. Because generation is bound by memory bandwidth, this ordering follows bandwidth rather than any gaming benchmark.

The smallest GPUs that still run Kosmos-2.5

The cheapest route in, by memory capacity. A tight fit runs the model but leaves nothing spare for a longer conversation.

What the numbers mean

What it takes to run this model

Minimum card

Tesla C1080

Memory needed

2.1 GB

Fastest

2,606 tok/s

Kosmos-2.5 is small enough at 1.3B parameters that hardware is rarely the obstacle — 818 of the cards we track can run it, including cards several years old.

The smallest card that holds it is the Tesla C1080 with 4 GB, running it at Q8_0 and producing around 28.4 tokens per second.

A B200 is the fastest we calculate for it: about 2,606 tokens per second, from 8,000 GB/s of memory bandwidth.

Background

Kosmos-2.5 was published by Microsoft, in United States of America, in August 2024. It comes out of industry.

It works in Multimodal, Language, Vision, and is recorded as doing character recognition (OCR), Document classification, Language modeling/generation, Visual question answering, Document representation.

It is derived from Pix2Struct-Large rather than trained from scratch, which is the usual way a specialised model is produced.

The weights being open is what puts this page in the calculator rather than only in the catalogue: it is a model you can actually hold. It is published under the microsoft organisation on Hugging Face.

Reading the throughput figures

Across every card that can run it, the middle of the range is about 73.2 tokens per second, and 797 of them clear the ten tokens per second that roughly matches reading speed.

Being dense, it reads all of itself per token, which is why the ordering by speed below follows the ordering by memory bandwidth so closely.

Its internal architecture is not on file, so memory is approximated from the parameter count and marked accordingly. Expect the real figure to differ, more so at long context.

What went into building it

Training it took roughly 2.2 × 10²¹ FLOP of computation — a measure of what producing the model cost, not of how fast it answers.

The training set ran to roughly 260,000,000,000 tokens.

Step by step

How to choose a GPU for Kosmos-2.5

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

01

Read the memory figure first

Look at what Kosmos-2.5 actually needs — around 2.1 GB at Q8_0. No amount of processing power compensates for a card that cannot hold it.
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 Kosmos-2.5 stops fitting a card that seemed fine.
03

Choose how far you will compress it

The quantisation column varies by card, because a bigger card holds a more accurate copy of Kosmos-2.5 — Q8_0 on the smallest card that fits. Set a floor to hold the comparison at one level.
04

Sort by speed

Ranking by tokens per second for Kosmos-2.5 follows memory bandwidth, not core counts, which is why the B200 tops it at 2,606 tok/s.
05

Look at the headroom, not just the fit

Tight means Kosmos-2.5 loads and works, with no room to raise the context later. Comfortable means you can. The difference matters more than a few tokens per second.
06

See what else that card runs

Every card name links to its own page, which runs the same calculation across the whole model catalogue. Worth a look before buying for Kosmos-2.5 alone — a card is usually bought for more than one model.

Answers

Kosmos-2.5 — common questions

Can I run Kosmos-2.5 on a 16 GB GPU?

Yes. A Tesla V100 SXM2 16 GB with 16 GB runs it at Q8_0, using about 2.1 GB and generating roughly 368 tokens per second — a comfortable fit.

Can I run Kosmos-2.5 on a 24 GB GPU?

Yes. A GeForce RTX 5090 D V2 with 24 GB runs it at Q8_0, using about 2.1 GB and generating roughly 437 tokens per second — a comfortable fit.

Is Kosmos-2.5 open source?

Its weights are published, so Kosmos-2.5 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.

How many parameters does Kosmos-2.5 have?

Kosmos-2.5 has 1.3B parameters. "KOSMOS-2.5 contains a total of 1.3 billion parameters" "The vision encoder is initialized from the Pix2Struct-Large model’s encoder (Lee et al. 2023), which is based on the Vision Transformer (ViT) (Dosovitskiy et al. 2021)". 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.

Who created Kosmos-2.5?

Kosmos-2.5 was published by Microsoft, based in United States of America, categorised as industry.

When was Kosmos-2.5 released?

Kosmos-2.5 was published in August 2024. Capability per parameter has improved considerably since, so a newer model of the same size is often the better use of the same hardware.

What is Kosmos-2.5 used for?

Kosmos-2.5 works in Multimodal, Language, Vision, and is recorded as handling character recognition (OCR), Document classification, Language modeling/generation, Visual question answering, Document representation. These are the areas it was designed around; they describe intent rather than a hard boundary.

Where can I download Kosmos-2.5?

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

How much compute was used to train Kosmos-2.5?

Around 2.2 × 10²¹ 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.

Can I run Kosmos-2.5 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 Kosmos-2.5 is rarely worth using. Every figure here assumes the whole model is on the card.

Would two GPUs run Kosmos-2.5 faster?

Two cards buy memory rather than speed. That matters for Kosmos-2.5 only if one card cannot hold it — 818 can, so a second adds little.

Why does the quantisation differ between cards for Kosmos-2.5?

A larger card holds a more accurate copy. Across the cards that run Kosmos-2.5, 1 compression levels are used; the floor control above pins it to one.

How accurate are these Kosmos-2.5 speed estimates?

These are estimates with real error bars. The fastest result here, 1,564–4,170 tok/s on the B200, could reasonably land anywhere in its published range depending on which runtime you use.

What GPU do I need to run Kosmos-2.5?

The smallest card in our catalogue that holds Kosmos-2.5 is the Tesla C1080, with 4 GB of memory. It runs the model at Q8_0 using about 2.1 GB, and produces roughly 28.4 tokens per second. 818 cards in total can run it.

How fast is Kosmos-2.5 on a GPU?

It depends on the card. The quickest we calculate is a B200 at about 2,606 tokens per second; the slowest that still runs it manages considerably less. Reading speed is around ten tokens per second, and 797 of the cards that can run Kosmos-2.5 clear that.

How much VRAM does Kosmos-2.5 need?

About 2.1 GB at Q8_0 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.

Can I run Kosmos-2.5 on a 8 GB GPU?

Yes. A CMP 170HX 8 GB with 8 GB runs it at Q8_0, using about 2.1 GB and generating roughly 485 tokens per second — a comfortable fit.

Can I run Kosmos-2.5 on a 12 GB GPU?

Yes. A GeForce RTX 3080 Ti with 12 GB runs it at Q8_0, using about 2.1 GB and generating roughly 297 tokens per second — a comfortable fit.

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.

All AI models