Typst in 2026

Author

Derek Sollberger

Published

July 13, 2026

Typst in 2026

Typst!
  • Introduce Typst for typesetting
  • Explore layouts
  • Updates from the past year
  • Reveal some works in progress

founders

Typst Team

inspiration

Laurenz’s blog post

  • compile in milliseconds (Rust)

  • data loading: XML, JSON, CSV

  • accessible documents by default

  • styling and layouts

  • scripting language

  • lightweight (about 40 MB)

  • secure binary

  • actively maintained

    • over 300 open-source volunteers

  • integration: CLI, Rust library

  • Typst app

  • Quarto

    • extensions
    • system2("quarto", c("add", "mcanouil/quarto-highlight-text@2.3.0"))
    • quarto::quarto_add_extension(extension = "mcanouil/quarto-highlight-text@2.3.0")

  • VS Code

  • Collabst collaboration

Syntax

  • [blank line for a paragraph break]
*strong face*
_emphasized text_
`verbatim text`
https://nyhackr.org/
#set heading(numbering: "1.")
#set heading(numbering: "1.a")

= Header 1
== Header 2
=== Header 3

+ ordered
+ list

- unordered
- list

- nested
  - list
${x}_(i=1)^n$
$sin(alpha + beta x)$
$display(lim_(n arrow.r oo)) frac(1, 1 + n)$
$ sum_(k=0)^n k
    &= 1 + ... + n  && "definition of summation" \
    &= (n(n+1)) / 2 && "proved by induction"$
    
#let year = 2026
$#year > 42$
$ mat(..#range(1, 17).chunks(4)) $
#math.equation(
  alt: "output equals weights times inputs plus bias",
  block: true,
  $ y = wx + b $,
)

Syllabus

Ever since Carlos Scheidegger’s talk at posit::conf(2023), I have wanted to try to use Typst to create a course syllabus. Over the past day, I have been aided by YouTube videos and the documentation on the Typst website to guide me into creating a fairly creative deliverable. In this blog post, I wanted to elaborate on some design decision and jot down some Typst code just in case I try to teach these materials in the future.

I should mention that for the past year, I have simply made straightforward syllabi in LaTex (via Overleaf). I later learned that my university has a welcoming culture that encourages creativity in teaching. For the first page, I realized that I didn’t need the official labels (course description, course learning outcomes) to appear immediately, so I relegated those to page two. Instead, I found libraries in the Typst Universe that allowed me to make diagrams—with the markup language!

  • treet: make file organization trees
  • fletcher: one (of several) libraries that make flow charts. I chose this one since the syntax was similar to Mermaid

I start out the syllabus with an orange #block, and containers in Typst still allow for markdown such as level-one headers. I proceed with more headers. Like Quarto markdown, syntax are treated as void tags (instead of needing to worry about opening and closing environments).

While Typst allows for creation of columns, the column settings tend to be too powerful as they pertain to all pages. One could make exceptions here and there.

Instead, like the documentation implied, I quickly preferred to use the #grid function for the layout.

#text(black, tree-list(
  marker: text(rgb("f58025"))[├── ],
  last-marker: text(rgb("f58025"))[└── ],
  indent: text(rgb("f58025"))[│#h(1.5em)],
  empty-indent: h(2em),
)[
  - Machine Learning
    - Supervised Learning
      - Regression
        - Penalization
      - Classification
        - Decision Trees
        - Random Forests
          - Boosting
        - SVMs
          - Kernels
    - Unsupervised Learning
      - Clustering
      - KNN
  - Neural Networks
    - Perceptrons
    - Backpropagtion
    - Image Processing
    
  - Large Language Models
    - Text Tokenization
    - Prompt Engineering
    - Developer Tools
]
#diagram(
    node-stroke: .1em,
    node-fill: gradient.radial(rgb("f58025").lighten(80%), rgb("f58025"), center: (30%, 20%), radius: 60%),
    spacing: 4em,

 node((0.5,1), "S", radius: 1.5em),
 node((1.5,1), "M", radius: 1.5em),
 node((2.5,1), "L", radius: 1.5em),
 
 node((0.5,2), "3", radius: 1.5em),
 node((1.5,2), "0", radius: 1.5em),
 node((2.5,2), "1", radius: 1.5em),
 
 node((0,3), "F", radius: 1em),
 node((1,3), "A", radius: 1em),
 node((2,3), "L", radius: 1em),
 node((3,3), "L", radius: 1em),
 
    node((0,4), "2", radius: 1em),
 node((1,4), "0", radius: 1em),
 node((2,4), "2", radius: 1em),
 node((3,4), "5", radius: 1em),

 edge((0.5,1), (0.5, 2), "-|>"),
 edge((0.5,1), (1.5, 2), "-|>"),
 edge((0.5,1), (2.5, 2), "-|>"),
 edge((1.5,1), (0.5, 2), "-|>"),
 edge((1.5,1), (1.5, 2), "-|>"),
 edge((1.5,1), (2.5, 2), "-|>"),
 edge((2.5,1), (0.5, 2), "-|>"),
 edge((2.5,1), (1.5, 2), "-|>"),
 edge((2.5,1), (2.5, 2), "-|>"),

 edge((0.5,2), (0, 3), "-|>"),
 edge((0.5,2), (1, 3), "-|>"),
 edge((0.5,2), (2, 3), "-|>"),
 edge((0.5,2), (3, 3), "-|>"),
 edge((1.5,2), (0, 3), "-|>"),
 edge((1.5,2), (1, 3), "-|>"),
 edge((1.5,2), (2, 3), "-|>"),
 edge((1.5,2), (3, 3), "-|>"),
 edge((2.5,2), (0, 3), "-|>"),
 edge((2.5,2), (1, 3), "-|>"),
 edge((2.5,2), (2, 3), "-|>"),
 edge((2.5,2), (3, 3), "-|>"),

 edge((0,3), (0, 4), "-|>"),
 edge((0,3), (1, 4), "-|>"),
 edge((0,3), (2, 4), "-|>"),
 edge((0,3), (3, 4), "-|>"),
 edge((1,3), (0, 4), "-|>"),
 edge((1,3), (1, 4), "-|>"),
 edge((1,3), (2, 4), "-|>"),
 edge((1,3), (3, 4), "-|>"),
 edge((2,3), (0, 4), "-|>"),
 edge((2,3), (1, 4), "-|>"),
 edge((2,3), (2, 4), "-|>"),
 edge((2,3), (3, 4), "-|>"),
 edge((3,3), (0, 4), "-|>"),
 edge((3,3), (1, 4), "-|>"),
 edge((3,3), (2, 4), "-|>"),
 edge((3,3), (3, 4), "-|>"),
)
#import "@preview/fletcher:0.5.8" as fletcher: diagram, node, edge
#import "@preview/showybox:2.0.4": showybox
#import "@preview/treet:1.0.0": *

#set page("us-letter") //thanks Europeans!
#set text(
  font: "New Computer Modern",
  size: 14pt
)

#set align(center)
#block(
  fill: rgb("f58025"),
  inset: 14pt,
  width: 100%,
  [= SML 301: Data Intelligence
= Modern Data Science Methods]
)

== Instructor: Derek Sollberger
=== (teacher email)

#set align(center)


#grid(
  columns: (1fr, 1fr),
  [== Post-Docs:

- person name (person email)
- person name (person email)],
  [== Teaching Assistants:

- person name (person email)
- person name (person email)
- person name (person email)
], [], [],
)

#set align(left+horizon)
#set text(
  font: "New Computer Modern",
  size: 12pt
)

#grid(
  columns: (1fr, 1fr),
  [
#text(black, tree-list(
  marker: text(rgb("f58025"))[├── ],
  last-marker: text(rgb("f58025"))[└── ],
  indent: text(rgb("f58025"))[│#h(1.5em)],
  empty-indent: h(2em),
)[...])
    
  ],
  [
#diagram(...)    
  ]
)

splash page

Now, we get to a page with a lot of text. I deploy a #grid to create two columns.

  • columns: (1fr, 4fr) computes that 20% of the width is allocated for the first column, and the remaining 80% is allocated for the second column
  • align: horizon aligns the content of the grid boxes with the vertical midpoint.

I also increase the font size to fill out space right up to before when an additional page would have been necessary.

#pagebreak()
// Course Description and CLOs
#set align(left)
#set text(
  font: "New Computer Modern",
  size: 14pt
)

#grid(
  align: horizon,
  columns: (1fr, 4fr),
  fill: (rgb("f58025"), none),
  gutter: 12pt,
  [],
  [== Course Description
  ...
  
  == Course Learning Outcomes
...

=== Prerequisites
...
]
)

page two

For the course reading list, I wanted to show off and combine images of some of the several books that I have read in the past couple of years. I start out by making a #grid with 6 rows and 5 columns (i.e. 4x3 matrix of images, and then some padding). The square() function literally draws a square, and I used that to draft the layout. Then, the image() function loads the images (where I have placed the images in an “image” subdirectory out of habit). Finally, I type out the book titles and authors in the padding for accessibility purposes. It is nice that text is automatically wrapped within grid cells.

== Reading List
// Grid of Suggested Books
#set text(
  font: "New Computer Modern",
  size: 10pt
)

#grid(
  align: center+horizon,
  columns: (1fr, 1fr, 1fr, 1fr, 1fr),
  rows: (1fr, 1fr, 1fr, 1fr, 1fr, 1fr),
  square(fill: rgb("f58025")),
  [_Deep Learning Illustrated_ \ Jon Krohn et al],
  [_ISLP_ \ James, Witten, Hastie, Tibshirani, Taylor],
  [_ML Cookbook_ \ Gallatin, Albon],
  square(fill: rgb("f58025")),
  square(fill: rgb("f58025")),
  image("images/Deep_Learning_Illustrated.png"),
  image("images/ISLP.png"),
  image("images/Machine_Learning_Cookbook.png"),
  square(fill: rgb("f58025")),
  [_DL with Python_ \ Francois Chollet],
  image("images/Deep_Learning_with_Python.png"),
  image("images/Hands_On_LLMs.png"),
  image("images/Statquest_Neural_Networks.png"),
  [_StatQuest AI_ \ Josh Starmer],
  [_Probabilistic ML_ \ Kevin P Murphy],
  image("images/PML.png"),
  image("images/Programmers_Brain.png"),
  image("images/Why_Machines_Learn.png"),
  [_Why Machines Learn_ \ Anil Ananthaswamy],
  [_Co-Intelligence_ \ Ethan Mollick],
  image("images/Co_Intelligence.png"),
  image("images/How_Data_Happened.png"),
  image("images/Weapons_of_Math_Destruction.png"),
  [_Weapons of Math Destruction_ \ Cathy O'Neil],
  square(fill: rgb("f58025")),
  [_Hands-On LLMs_ \ Allammar, Grootendorst],
  [_How Data Happened_ \ Wiggins, Jones],
  [_Programmer's Brain_ \ Felienne Hermans],
  square(fill: rgb("f58025")),
)

mosaic

What finally sold me on Typst was the notion that I would not need “yet another markdown language” for making tables. In fact, Typst is more than a markup language. Here, we literally #let a data file exist as an object in a variable name, and then print the variable with a for loop.

For further appeal, if need be, the tables

  • can be rotated (say, to landscape mode)
  • can be set to automatically split themselves across pages #show figure: set block(breakable: true)
== Lecture Schedule
// Table of Lecture Sessions (one column)

#set text(
  font: "New Computer Modern",
  size: 10pt
)

//load schedule data
#let lecture_schedule = csv("SML_301_Fall_2025_lecture_schedule.csv", row-type: array).slice(2)

//header row bold
#show table.cell.where(y: 0): set text(weight: "bold")

//allow table to be split across pages!
#show figure: set block(breakable: true)

//main table code
#table(
  align: center,
  columns: 7,
  table.header[Week][Date][Topic][Python][Matplotlib][Precept][Projects],
  ..for(.., Week, Date, Topic, Python, Matplotlib, Precept, Projects) in lecture_schedule{
    (Week, Date, Topic, Python, Matplotlib, Precept, Projects)
  },
  fill: (none, none, rgb("f58025"), none, none, none, rgb("f58025")),
)

table

You may have noticed that I reduced the font size for materials that are of less importance to the reader. That is, there are some sections that I do not need students to read right away, but at the same time I want to assure them that these matters are available in the syllabus.

#set text(
  font: "New Computer Modern",
  size: 10pt
)

#grid(
  align: horizon,
  columns: (1fr, 4fr),
  fill: (rgb("f58025"), none),
  gutter: 12pt,
  [],
  [= Class Policies
- Lecture sections: ...
- Precepts ...
- Late policy: ...
- Computers: ...
- Special Accommodations:  ...
- Academic Integrity: ...
],
)

policies

In contrast, there are some unique materials in my syllabi. In Quarto, I love using callout boxes (as they remind me of textbooks from when I was a student). Here in the Typst Universe, the showybox library is the analogue. I quickly adapt its syntax while also being mindful of colors and readability.

Also, footnote() is still as convenient as its LaTeX counterpart.

#showybox(...)

#showybox(...)

#showybox(
  frame: (
    //border-color: red.darken(50%),
    title-color: rgb("f58025"),
    body-color: rgb("f58025")
  ),
  title-style: (
    color: black,
    weight: "bold",
    align: center
  ),
  shadow: (
    offset: 5pt,
  ),
  title: "Typst",
  [This syllabus document was created with Typst!
  https://typst.app
  ], 
)

take home

Recent Patches

Accessibility

#figure(
  stack(
    dir: ltr,
    spacing: 0.5em,
    rect[Tagged PDF],
    text(2em, sym.arrow.long),
    rect[Accessibility],
  ),
  alt: "Diagram with two rectangles. The first is labelled 'Tagged PDF'. An arrow points to the second, labelled 'Accessibility'",
  caption: [
    Tags enable PDF accessibility
  ],
)

tagged PDF

PDFs standards

PDFs standards
  • image source: Typst app
  • Patch 0.15: multiple PDF standards simultaneously

PDFs as Images

#figure(
  image(
    "world_cup_semis.pdf",
    alt: "Semi final bracket for World Cup 2026: France, Spain, England, Argentina"
  ),
  caption: [
    Semi-final bracket for World Cup 2026: France, Spain, England, Argentina
  ],
)

tagged PDFs as Images

Character-Level Justification

#set par(
  justify: true,
  leading: 0.50em,
)
#lorem(256)

character-level justification

Typst uses the same fundamental algorithm as LaTeX does to optimize paragraphs … as character-level justification is a feature that LaTeX does not support.

Variable Fonts

#set text(font: "Lora")

#for weight in range(
  400,
  700,
  step: 50,
  inclusive: true,
) [
  #text(weight: weight)[All this has happened before,]
  \
]

variable fonts
  • remember to mouseover the text(font: "Lora") setting to see available variations through a tooltip

native support for equations with MathML

  • selectable text
  • screen readers
  • high resolution for browsers
  • experimental (beta)
#document("index.html", title: [Home])[
  #title()
  #link(<blog>)[Go to blog]
]

#document("blog.html", title: [Blog])[
  #title()
  Welcome to my blog!
] <blog>

#asset("robots.txt", "Disallow: *")
  • by-chapter bibliographies
  • by-topic bibliographies

Works in Progress

Flash Cards

#set text(
  font: "New Computer Modern",
  size: 36pt
)
#set page(paper: "us-letter", 
          margin: (x: 1cm, y: 1cm), )
#set align(center+horizon)

#grid(
  columns: (1fr, 1fr, 1fr),
  rows: (1fr, 1fr, 1fr),
  [Dwight \ Schrute],
  [Creed \ Bratton],
  [Angela \ Kinsey],
  [Meredith \ Palmer],
  [Jim \ Halpert],
  [Kevin \ Malone],
  [Stanley \ Hudson],
  [Phyllis \ Vance],
  [Pam \ Beesley],
)
#set align(center+horizon)
#grid(
  columns: (1fr, 1fr, 1fr),
  rows: (1fr, 1fr, 1fr),
  gutter: 5pt,
  [#image("angela_kinsey.png")],
  [#image("creed_bratton.png")],
  [#image("dwight_schrute.png")],
  [#image("kevin_malone.png")],
  [#image("jim_halpert.png")],
  [#image("meredith_palmer.png")],
  [#image("pam_beesley.png")],
  [#image("phyllis_vance.png")],
  [#image("stanley_hudson.png")],
)

“student” names

“student” pictures

MCQ Organizer

#set enum(numbering: "1).A)")

+ Which employee is known for his chili?
  + Kevin Malone
  + Creed Bratton
  + Ryan Howard
  + Andy Bernard
  + Darryl Philbin

+ Which employee is known for his "thoughts" blog?
  + Creed Bratton
  + Kevin Malone
  + Ryan Malone
  + Andy Bernard
  + Darryl Philbin
  
#let mcq_example = csv("mcq_example.csv")

#show table.cell.where(y: 0): set text(weight: "bold")

#rotate(
  -90deg,
  reflow: true,
table(
  align: center,
  columns: 6,
  //table.header[Question][Correct][Alt2][Alt3][Alt4][Alt5],
  ..for(.., Question, Correct, alt2, alt3, alt4, alt5) in mcq_example{
    (Question, Correct, alt2, alt3, alt4, alt5)
  },
  fill: (rgb("f58025"), rgb("f58025"), none, none , none, none),
)
)

MCQ structured

workflow

Interactive Machine Learning

Ridge Regression

tidymodels
  • \(\lambda = 0.1:\)

\[\hat{y} = 7.550 - 0.292X_{1} - 0.678X_{2} - 0.274X_{3}\]

  • \(\lambda = 1.0:\)

\[\hat{y} = 7.840 - 0.085X_{1} + 0.132X_{2} - 0.135X_{3}\]

  • Typst and Marimo?

Music Notes

staves

guitar tabs

arrangement

Causal Graphs

Using the autograph package in Typst

directed acyclic graph

chain

confounder

collider

Evidence of Attention

attention example
#set text(fill: rgb("#AAAAAA"),
weight: 400)

In #text(fill: red.lighten(50%), size: 1.5em, weight: 500)[machine learning], #text(fill: red, size: 2.0em, weight: 700)[attention] is a method that determines the #text(fill: red.lighten(60%), size: 1.5em, weight: 550)[importance] of each component in a #text(fill: red.lighten(70%), size: 1.75em, weight: 600)[sequence] relative to the other components in that #text(fill: red.lighten(70%), size: 1.75em, weight: 600)[sequence]. In #text(fill: red.lighten(85%), size: 1.9em, weight: 650)[natural language processing], #text(fill: red.lighten(60%), size: 1.5em, weight: 550)[importance] is represented by "soft" weights assigned to each word in a sentence. More generally, #text(fill: red, size: 2.0em, weight: 700)[attention] encodes vectors called token embeddings across a fixed-width #text(fill: red.lighten(70%), size: 1.75em, weight: 600)[sequence] that can range from tens to millions of tokens in size.

---Wikipedia
  • color gradients
  • variable fonts
  • scripting language
  • in interactive production?

Coda

Donald Knuth on TeX

thanks!
sessionInfo()
R version 4.6.1 (2026-06-24 ucrt)
Platform: x86_64-w64-mingw32/x64
Running under: Windows 10 x64 (build 19045)

Matrix products: default
  LAPACK version 3.12.1

locale:
[1] LC_COLLATE=English_United States.utf8 
[2] LC_CTYPE=English_United States.utf8   
[3] LC_MONETARY=English_United States.utf8
[4] LC_NUMERIC=C                          
[5] LC_TIME=English_United States.utf8    

time zone: America/New_York
tzcode source: internal

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

loaded via a namespace (and not attached):
 [1] htmlwidgets_1.6.4 compiler_4.6.1    fastmap_1.2.0     cli_3.6.6        
 [5] tools_4.6.1       htmltools_0.5.9   otel_0.2.0        rstudioapi_0.19.0
 [9] yaml_2.3.12       rmarkdown_2.31    knitr_1.51        jsonlite_2.0.0   
[13] xfun_0.57         digest_0.6.39     rlang_1.2.0       evaluate_1.0.5