A fast DuckDB scan can still stall inside a Python UDF.
Read: Python UDFs at Native Speed? exaloop.io/blog/python-udfs-…
32
From Python research to parallel execution.
One billion price updates: 19 minutes in Python, 3.8 seconds on a GPU with Codon. Same simulation logic; different execution target. See the code and benchmark setup. Read From Quant Research to Parallel CPU and GPU Without Rewriting the Strategy →
exaloop.io/blog/quant-cpu-gp…
Your GPU may not be the bottleneck. The Python feeding it might be.
Read What Happens When LLM-Generated Python Has to Go to Production? → exaloop.io/blog/when-llm-gen…
A lot of ML systems already rely on fast native kernels. The slower part can be the application-specific code around them: parsing, filtering, string work, dictionary lookups, feature extraction, batching, and business rules.
Coding agents make that custom Python cheaper to generate—but not necessarily cheaper to run in production.
In this article, we start with an LLM-generated preprocessing pipeline, profile the end-to-end system, try the obvious Python fixes first, and then compile the CPU-heavy region with Codon. We compare throughput, CPU, memory, latency, GPU utilization, and how much code actually had to change.
The point isn’t “Python is slow.” It’s to find out what the profiler says—and what the least disruptive production path looks like.
Codon can now be used on Compiler Explorer! exaloop.io/blog/codon-ce
1
62
We just released Codon 0.19 🚀 Learn more in our blog post: exaloop.io/blog/codon-019
1
60
Latest Codon news, features, and what we have in the works!
linkedin.com/pulse/codon-loo…
1
50
We're thrilled to finally be releasing a Codon-native, compiler-accelerated NumPy implementation, and also share that Codon is now truly open source under the Apache-2.0 license. Learn more in our announcement blog post: exaloop.io/blog/codon-2025
Check out the recent MIT News article about Codon! news.mit.edu/2023/codon-pyth…
Check out our first blog post about Codon internals, "Mapping Python to LLVM": blog.exaloop.io/python-llvm/
Exaloop retweeted
Codon: A high-performance Python compiler using LLVM
#python #llvm #n #z #exaloopcodon #codon #release #zerooverhead #compiler #b #extensible #gpu #highperformance #run #x #using
github.com/exaloop/codon
We're glad to finally release Codon, a high-performance, extensible #Python compiler that uses LLVM to generate optimized native code, typically gets 10-100x speedups over Python, and supports multithreading and GPU for speedups many times higher still. github.com/exaloop/codon





