GLM-4.6 TPS calculator

Open weights Z.ai (Zhipu AI),Tsinghua University 357B parameters September 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

4 cards that can run it

818 cards we hold specifications for

Smallest card that fits

Radeon Instinct MI325X

256 GB · Q4_K_M · 12.8 tok/s

Fastest card

B300

17.0 tok/s · 288 GB

Which GPUs can run GLM-4.6?

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.

4 cards match

Calculating
Needs Quantisation Fit
17.0 tok/s

10–27 · low confidence

B300 NVIDIA 288 GB 8,000 GB/s Sep 2025 258.2 GB Q5_K_M Tight
13.5 tok/s

8–22 · low confidence

Radeon Instinct MI350X AMD 288 GB 8,190 GB/s Jan 2025 258.2 GB Q5_K_M Tight
13.5 tok/s

8–22 · low confidence

Radeon Instinct MI355X AMD 288 GB 8,190 GB/s Jan 2025 258.2 GB Q5_K_M Tight
12.8 tok/s

8–21 · low confidence

Radeon Instinct MI325X AMD 256 GB 6,000 GB/s Oct 2024 216.7 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
Z.ai (Zhipu AI),Tsinghua University
Organisation type
Industry,Academia
Country
China
Published
30 September 2025

What it does

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

Domain
Language
Task
Language modeling/generation, Question answering, Code generation, Quantitative reasoning, System control

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
357B

Similarly to GLM 4.5: 355 billion total parameters (reported) with 32 billion active parameters (assumed)

Training data
23,000,000,000,000 tokens

23T tokens (from Jaime's correspondence with the GLM team)

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
4.4 × 10²⁴ FLOP

6 FLOP/parameter/token * 32000000000 active parameters [very likely assumption - everything else is reported to be same as at GLM 4.5] * 23000000000000 tokens = 4.42e24 FLOP

How it was established
Operation counting

The training run

What it physically took to train: which chips, how many, for how long, and what that drew from the wall.

Wall-clock time
2,880 hours (120 days)

4 months (from Jaime's correspondence with the GLM team) 4 months * 30 days * 24 hours = 2880 hours

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

MIT license https://huggingface.co/zai-org/GLM-4.6

Hugging Face
zai-org

How it is classified

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

Why it is tracked
Discretionary

"We evaluated GLM-4.6 across eight public benchmarks covering agents, reasoning, and coding. Results show clear gains over GLM-4.5, with GLM-4.6 also holding competitive advantages over leading domestic and international models such as DeepSeek-V3.1-Terminus and Claude Sonnet 4."

Record confidence
Likely

Sources

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

Reference
GLM-4.6: Advanced Agentic, Reasoning and Coding Capabilities
Last updated
30 December 2025

The extremes

The ten fastest GPUs that run GLM-4.6

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.

  1. 01 B300 288 GB · 8,000 GB/s · Q5_K_M 17.0 tok/s
  2. 02 Radeon Instinct MI350X 288 GB · 8,190 GB/s · Q5_K_M 13.5 tok/s
  3. 03 Radeon Instinct MI355X 288 GB · 8,190 GB/s · Q5_K_M 13.5 tok/s
  4. 04 Radeon Instinct MI325X 256 GB · 6,000 GB/s · Q4_K_M 12.8 tok/s

What the numbers mean

What it takes to run this model

Minimum card

Radeon Instinct MI325X

Memory needed

216.7 GB

Fastest

17.0 tok/s

GLM-4.6 reaches a parameter count of 357B. That is beyond what any single graphics card holds. Running it means either splitting it across several cards or renting hardware built for the job, and every card able to hold it alone is a datacentre part. The number that can: 4.

The smallest card that holds it is Radeon Instinct MI325X, with a memory capacity of 256 GB, running it at a compression of Q4_K_M and producing around 12.8 tokens per second.

Top of the range is B300, generating roughly 17.0 tokens per second on the strength of a memory bandwidth of 8,000 GB/s.

What this model is

GLM-4.6 was published by Z.ai (Zhipu AI),Tsinghua University, in the country recorded as China, during September 2025. The category the publisher falls under is industry,Academia.

It works in the domain of Language, and is recorded as performing the task of language modeling/generation, Question answering, Code generation, Quantitative reasoning, System control.

Published weights mean the model runs on your machine rather than someone else's, which is what makes the hardware question below answerable at all. On Hugging Face it is published under the organisation zai-org.

What decides the speed

Across every card that can run it, the middle of the range sits at 13.5 tokens per second. Exceeding reading speed outright: 4 of them.

Every weight participates in every token here, so bandwidth is the whole story: the ranking below is effectively a ranking of memory throughput.

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.

Training and provenance

The training run consumed about 4.4 × 10²⁴ FLOP. That figure measures what producing the model cost, and has no bearing on how fast it answers.

Training consumed a corpus of around 23,000,000,000,000 tokens of text.

It is tracked in the underlying dataset for one reason in particular: discretionary.

Step by step

How to choose a GPU for GLM-4.6

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

    Start from what it actually needs, which is the requirement of GLM-4.6, needing around 216.7 GB at a compression of Q4_K_M. That figure, not the headline performance of a card, is what decides whether it runs.

  2. 02

    Match the context to your actual use

    Longer conversations cost memory on top of what the weights need. Move the slider to your real working length before trusting any row for GLM-4.6.

  3. 03

    Choose how far you will compress it

    Compression is what makes a model fit smaller cards, at some cost in accuracy, reaching a compression of Q4_K_M on the smallest card that fits. Setting a minimum quality drops the cards that only manage it by squeezing further than you would want, and holds the comparison at one level.

  4. 04

    Sort by speed

    Sort by speed to see how cards rank for GLM-4.6. It will not match a gaming ordering, because generation is bound by memory bandwidth. The card topping the list is B300, at 17.0 tok/s.

  5. 05

    Look at the headroom, not just the fit

    A tight fit runs, but leaves nothing spare for a longer conversation, in the case of GLM-4.6. Comfortable means you can grow the context later. That difference matters more than a few tokens per second, so buy for comfortable if you expect to.

  6. 06

    Check the card from the other side

    Following a card through to its own page shows every other model it can hold, which is the question that follows once you have settled on GLM-4.6.

Answers

GLM-4.6 — common questions

01

GLM-4.6— can I run it if it does not fit in my GPU?

It can be split between the card and system memory, but it generates painfully slowly that way. The nearest miss we calculate falls short by 43.9 GB. Every figure here assumes the whole model is resident on the card.

02

GLM-4.6— would two GPUs run it faster?

A second card roughly doubles the memory available but not the generation rate. The number already able to run it alone: 4. So a second card is rarely the answer here.

03

GLM-4.6— why does the quantisation differ between cards?

Each card is shown running the least-compressed copy it can hold. The number of distinct compression levels across the cards that fit it: 2. Bigger cards get the more accurate version, and the quality floor above pins the comparison to one level.

04

GLM-4.6— how accurate are these speed estimates?

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: 10–27 tok/s on B300. The same model and card vary by thirty to fifty per cent depending on the inference software and its version.

05

GLM-4.6— what GPU do I need to run it?

The smallest card in our catalogue that holds it is Radeon Instinct MI325X, with a memory capacity of 256 GB. It runs the model at a compression of Q4_K_M using about 216.7 GB, and produces roughly 12.8 tokens per second. The number of cards able to run it in total: 4.

06

GLM-4.6— how fast is it on a GPU?

It depends on the card. The quickest we calculate is B300, at about 17.0 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: 4.

07

GLM-4.6— how much VRAM does it need?

It needs about 216.7 GB at a compression of Q4_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.

08

GLM-4.6— is it open source?

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.

09

GLM-4.6— how many parameters does it have?

It has a parameter count of 357B. Similarly to GLM 4.5: 355 billion total parameters (reported) with 32 billion active parameters (assumed). 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.

10

GLM-4.6— who created it?

It was published by Z.ai (Zhipu AI),Tsinghua University, based in China, an organisation categorised as industry,Academia.

11

GLM-4.6— when was it released?

It was published in September 2025.

12

GLM-4.6— what is it used for?

It works in the domain of Language, and is recorded as handling the task of language modeling/generation, Question answering, Code generation, Quantitative reasoning, System control. 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.

13

GLM-4.6— where can I download it?

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

14

GLM-4.6— how much compute was used to train it?

Training consumed around 4.4 × 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.

Source

Original publication

Record last updated 30 December 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.