Modverse #54: From GTC to Edinburgh, a Community Building Momentum

This edition covers one of the busiest stretches in Modular's recent history: four days at GTC, a new office on another continent, fresh community builds, and a release that expands what MAX and Mojo🔥 can do. Here's everything that's been happening across the ecosystem.

Community Innovations

This month’s spotlight:

• mojo-js: A Mojo-to-JavaScript Transpiler Experiment: Jeroen van Bemmel (jbemmel) built a proof-of-concept transpiler that takes Mojo-flavored source code and maps it to executable JavaScript, including a custom lexer, AST parser, and ~50 CI-tested cases. The goal is conceptual: exploring Mojo as an authoring language for the browser runtime. Demo · GitHub
• Practical Mojo Examples: ahmetax has been building and sharing a growing collection of 50+ tested Mojo 0.26.1 scripts on GitHub, a useful hands-on reference for developers learning the language. GitHub
• ui-terminal-mojo updated to 26.2.0: rd4com updated their pure-Mojo TUI framework to the 26.2.0 nightly, now confirmed working on macOS M4 and Linux with around 6MB RAM and 0.3 CPU usage. The framework uses an immediate-mode loop instead of callbacks, with auto-scaling dynamic layouts and a toggleable tab menu. No extra dependencies, just external_call. GitHub
• Marrow: Apache Arrow in Mojo: kszucs is building a native Apache Arrow implementation in Mojo, and it's already turning heads. The core abstractions are in place: arrays, builders, compute kernels, Python bindings, and zero-copy interop with PyArrow via the C Data Interface. Early benchmarks show Marrow running 1.3-3.9x faster than PyArrow for numeric, string, and nested list type conversions, with experimental GPU support via Mojo's DeviceContext. Arrow powers Pandas 2.0, Polars, Spark, and virtually every modern data tool. Mojo now has a path into that ecosystem. GitHub

• Decimo v0.9.0: CLI Calculator and Python Bindings: Yuhao Zhu shipped Decimo v0.9.0 (compatible with Mojo 26.2), and this release is a fun one. Decimo is now also a command-line arbitrary-precision calculator built on ArgMojo, with a tokenizer, shunting-yard parser, and RPN evaluator, all compiled to a single native binary with zero runtime dependencies. The release also adds Python bindings via PythonModuleBuilder, exposing BigDecimal as a native CPython extension, available on PyPI. pixi add decimo GitHub
• Mojo and the Missing Layer in Heterogeneous Compute: Jeroen van Bemmel followed up his mojo-js work with a discussion on the gap between programmable and economically addressable heterogeneous compute, and extended the project to ship Mojo-originated work to the Distributive Compute Platform (DCP) from the browser. A thought-provoking read on what a coordination layer between Mojo runtimes and distributed compute markets could look like. Prototype
• BlazeSeq, Mohamed Mabrouk's high-throughput FASTQ parser for bioinformatics pipelines, was featured at our March community meeting. BlazeSeq achieves 5+ GB/s single-threaded zero-copy parsing, includes a GPU-friendly Structure-of-Arrays batch API for direct host to device transfers, and integrates parallel gzip decompression via rapidgzip for up to 5x end-to-end speedup over standard tools. Python bindings are available via pip install blazeseq. Forum

💡 We love seeing what you create! Share your MAX or Mojo projects in the forum under the Community Showcase category so we can highlight them in the next edition.

Modular Making Waves

• We were at GTC! We came to GTC to show what's possible when you build AI infrastructure from the ground up. Over four days we ran live demos, met passionate engineers and researchers, and learned as much as we shared. Thank you to everyone who stopped by! There's a lot more to come. Everything we showed at GTC is available here.

• At GTC, we sat down with Lambda. Modular CEO and co-founder, Chris Lattner, and Lambda’s Sam Khosroshahi on how Modular and Lambda are shaping the future of AI infra. Full conversation dropping soon on Lambda's YouTube.
• Modular is officially in Edinburgh 🏴󠁧󠁢󠁳󠁣󠁴󠁿. We're at the Bayes Centre, where world-leading data science and AI teams work alongside businesses to turn research into real solutions. A fitting place to build the next layer of AI infrastructure. We’d love for you to join us! See our open roles.

• Sources News: "Trying to break Nvidia's lock-in." Alex Heath sat down with Chris Lattner on the GTC show floor for a conversation about Modular's bid to crack the software moat keeping the AI industry tethered to one chipmaker. Read it here.

Open Source Contributions

If you've recently had your first PR merged, message Inaara Walji in the forum to claim your Modular swag! Check out the recently merged contributions from our amazing community members:

• avijaira [1]
• bertaveira [1]
• bgreni [1]
• brian-carroll [1]
• byungchul-sqzb [1]
• DerekVeit [1]
• elpres [1]
• JosephWakeling [1]
• josiahls [1]
• kkimmk [1]
• KrxGu [1]
• lamija-i [1]
• mahendrarathore1742 [1] [2]
• martinvuyk [1]
• msaelices [1]
• nejraberberovic [1]
• pei0033 [1]
• prsabahrami [1]
• sbrunk [1]
• tae-su-kim [1]
• tolgacangoz [1]
• turakz [1]
• ulmentflam [1]
• walkerbrown [1]

Modular News & Events: Stay Connected

• Modular 26.2 is out. The latest release expands the Modular Platform to include image generation and editing workflows, with FLUX.2 model variants supported at over 4x speedup over state-of-the-art. On the Mojo side, 26.2 introduces new language features and AI coding skills purpose-built for GPU kernel development. Agents like Claude, Cursor, and Codex can now write, port, and optimize kernels across hardware targets using Mojo's Python-like syntax and 750K+ lines of open-source kernel code as context. Read the release blog.
• Missed our March community meeting? Mohamed Mabrouk presented BlazeSeq, Lukas from Modular did a deep dive into Variadic Metaprogramming in Mojo, and we walked through the 26.2 release highlights. Watch here.
• Structured Mojo Kernels series. If you haven't been following this series, now's a good time to catch up. Three posts in, the team walks through the architecture behind Modular's highest-performance kernels, from the design philosophy to TileIO, TilePipeline, and TileOp, through to how conv2d and block-scaled matmul are built on a shared foundation. Read Part 1.
• Programming Massively Parallel Processors is the gold standard GPU programming textbook and the 5th edition just dropped. Mojo solutions to every exercise are available. Check out the repo and see what GPU programming looks like in Mojo.