37 stories tagged code-generation, newest first.
160 points by rzk | 56 comments
GitHub repo with code and paper for GVS5H, a training-free self-orchestration harness letting Qwen3.8-27B instances match Claude Fable 5 on hard LiveCodeBench problems.
79 points by sumnole | 102 comments
OpenAI research publication on improving its models' coding abilities in low-resource programming languages like COBOL, Fortran, OCaml, and Visual Basic.
229 points by __cayenne__ | 103 comments
StarSkirmish Bench, a benchmark where LLMs play StarCraft: Brood War by writing code, showing gameplay and results for flagship models.
67 points by lasu | 13 comments
Parsimony is an open-source benchmark and leaderboard measuring how much net code AI models add when solving tasks, counting syntax-level coding units from published benchmark patches.
11 points by bradleygibbs | 7 comments
Blog post by Bradley Gibbs on designing AI-assisted development loops to qualify generated code for human review, reduce wasted work, and measure verification value.
53 points by mpweiher | 3 comments
Blog post describing how the rat compiler backend replaced its linear scan register allocator with a simpler priority bin-packing allocator for x86-64.
109 points by nerdypepper | 44 comments
Tangled repository page for bez, a project generating a browser engine from web specifications and test suites, showing files and recent commits.
62 points by thefourthchime | 38 comments
PacBench, a benchmark site comparing how well AI model and harness combinations recreate a Pac-Man game in a single HTML page from one prompt.
334 points by danielvaughn | 178 comments
Blog post introducing Huzzah, an experimental editor where you write pseudocode that is synced to real source code by AI on save, with the pseudocode kept as intent.
519 points by EvanZhouDev | 243 comments
Microsoft AI's announcement page introducing MAI-Code-1-Flash, a coding model for quick fixes and complex engineering tasks, available to try in VS Code.
900 points by mfiguiere | 417 comments
Qwen team blog post announcing the open-source release of Qwen3.6-27B, a dense 27B-parameter model claiming flagship-level coding performance.
605 points by Anon84 | 182 comments
arXiv paper showing that fine-tuning an LLM on its own sampled code outputs (simple self-distillation) improves code generation, e.g. Qwen3-30B on LiveCodeBench.
411 points by dnw | 368 comments
Substack essay from Vagabond Research arguing that LLMs produce plausible rather than correct code, using a simple database primary key lookup test as an example.
776 points by kachapopopow | 278 comments
Blog post by Can Bölük arguing that changing only the edit tool in a coding agent harness improved coding performance across 15 LLMs.
414 points by voxadam | 658 comments
IEEE Spectrum guest article by Jamie Twiss arguing newer AI coding assistants are more prone to silent but deadly failure modes.
590 points by damethos | 164 comments
A blog post that tongue-in-cheek implements an even/odd check using 4 billion generated if statements, inspired by a viral TikTok screenshot of beginner code.
398 points by samrolken | 283 comments
GitHub repository for nokode, an experimental web app where every HTTP request is handled directly by an LLM using database, response, and memory tools instead of code.
410 points by simonpure | 173 comments
Project page for LL3M, a University of Chicago research system where large language models write Blender code to create and iteratively edit 3D models.
![Screenshot of arxiv.org: GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models [pdf]](https://hn.urlbox.com/shots/renders/2025/8/12/68eb07e3-6f81-44a7-9646-4891d3adb444.jpg)
382 points by SerCe | 71 comments
arXiv technical report from Zhipu AI and Tsinghua introducing GLM-4.5, a 355B-parameter open-source MoE language model for agentic, reasoning, and coding tasks.
544 points by simonw | 362 comments
Simon Willison blog post showing the GLM-4.5 Air model, 3-bit quantized for MLX, generating a working Space Invaders game in JavaScript on a 64GB laptop.
680 points by danielhanchen | 300 comments
Qwen Team blog post announcing Qwen3-Coder, their most agentic code model to date, with links to GitHub, Hugging Face, ModelScope and Discord.
557 points by PaulHoule | 230 comments
arXiv paper from Inception Labs presenting Mercury, diffusion-based large language models for coding that generate multiple tokens in parallel at very high throughput.
496 points by nomilk | 315 comments
A Cursor forum bug report where a user says the AI editor refused to generate more code past about 800 lines, telling them to learn coding themselves.
331 points by joshwa | 113 comments
arXiv paper introducing SWE-Bench+, an empirical analysis revealing data quality issues in the SWE-bench coding benchmark for LLMs, such as solution leakage and weak tests.
409 points by lolptdr | 130 comments
Harper Reed's blog post describing his workflow for building software with LLMs: brainstorm a spec, plan, then execute with LLM codegen in discrete loops.
710 points by rcarmo | 411 comments
Max Woolf's blog post experimenting with whether repeatedly prompting an LLM to "write better code" actually improves the code it generates.
451 points by tosh | 133 comments
Mistral AI announcement of Codestral Mamba, a Mamba2-based language model specialized in code generation, released under the Apache 2.0 license.
418 points by alexmolas | 185 comments
Mistral AI blog post from May 29, 2024 announcing Codestral, its generative AI model for code.
398 points by ridruejo | 99 comments
LWN article on Cranelift, the Wasmtime code-generation backend now available as an optional faster debug-build backend for the Rust compiler.
581 points by rushingcreek | 272 comments
Phind blog post announcing Phind-70B, a code-focused language model claimed to approach GPT-4 Turbo code quality while running 4x faster.
448 points by albert_e | 233 comments
A post from Meta AI on X announcing the release of Code Llama 70B, a larger version of its code-generation language model.
604 points by Tomte | 114 comments
A page from the Everything curl book explaining the --libcurl option, which generates a C program using libcurl that replicates a curl command-line transfer.
1431 points by r4um | 451 comments
Blog post humorously exploring writing an even/odd checker using 4 billion if statements, inspired by a viral TikTok screenshot of beginner code.
602 points by nickthegreek | 236 comments
Meta newsroom post from August 24, 2023 announcing Code Llama, a large language model for generating and discussing code.
911 points by marcopicentini | 74 comments
Meta AI blog post from August 24, 2023 announcing Code Llama, a large language model for code generation built on Llama 2.
460 points by tlarkworthy | 113 comments
GitHub repository for smol-ai's Smol Developer, an open-source AI agent that generates entire codebases from prompts using large context windows.
335 points by mlejva | 162 comments
GitHub repository for e2b, an open-source IDE where developers write documentation and AI agents with tool access do the coding.