We trained a 32B model to beat Opus 4 at credit card optimization
We built an RL environment for credit card reward optimization and trained Qwen 32B with GRPO against it. The trained model scores ~0.51 on held-out tasks vs. Opus 4 at ~0.41 and GPT-4o at 0.36. Environment is open source (Apache 2.0). Blog post explains the reward design, what broke during training, how we fixed it, and what we'd do differently.
We trained a 32B model to beat Opus 4 at credit card optimization launched on April 20, 2026 with 6 votes, #1,677 of 21,248 launches that month and more than 75% of that year's launches.
What it does
Enter your monthly spending amounts, categories, and any card limits or preferences. The tool calculates the best mix of credit cards that maximizes your rewards and shows the expected cash‑back va...
Does a similar job
all alternatives →
Terminal-Bench-RL: Training long-horizon terminal agents with RL2025 · github.com · ▲125After training calculator agent via RL, I really wanted to go bigger! So I built RL infrastructure for training long-horizon terminal/coding agents that scales from 2x A100s to 32x H100s (~$1M worth of compute!) Without any training, my 32B agent hit #19 on Terminal-Bench leaderboard, beating Stanford's Terminus-Qwen3-235B-A22! With training... well, too expensive, but I bet the results would be good! *What I did*: - Created a Claude Code-inspired agent (system msg + tools) - Built Docker-isolated GRPO training where each rollout gets its own container - Developed a multi-agent…
- IBI built a tool to manage and compare credit card rewards2025 · rewards.getonecard.io · ▲87
This is a free tool that helps you manage and visualize your credit card rewards across different categories. You can input the cards in your wallet and see how they complement each other, spot gaps in your setup, and also see the best card to use for a given merchant. I’m also a founder at OneCard, where we’re building a smart card that’ll eventually handle all of this automatically, routing each purchase to the best card in real-time. Would love feedback from the HN community!
- AGA GPU-accelerated binary vector index2025 · rlafuente.com · ▲65
This is a vector index I built that supports insertion and k-nearest neighbors (k-NN) querying, optimized for GPUs. It operates entirely in CUDA and can process queries on half a billion vectors in under 200 milliseconds. The codebase is structured as a standalone library with an HTTP API for remote access. It’s intended for high-performance search tasks—think similarity search, AI model retrieval, or reinforcement learning replay buffers. The codebase is located at https://github.com/rodlaf/BinaryGPUIndex.
Minimal LLM Post-Training Experiments on an 8GB GPU (SFT, DPO, GRPO)Aug 2026 · github.com · ▲21Minimal, readable LLM post-training experiments on one 8GB GPU. Measures forgetting, seed variance, and RL emergence. - pochenai/nano-llm-posttraining
Supercharging RL with Hyper-Efficient Online Opt, +165% in 2h, $102025 · arc.computer · ▲32Arc enables continual learning for agents to adapt, learn, and transfer knowledge in live environments. Open-source SDK for building agents that learn from experience and improve over time.
Running Gemma-4 26B at 124 tokens/SEC on a CPU, no GPUJun 2026 · apeg.dev · ▲10I wanted to know how fast a 26B mixture-of-experts model could run on a desktop CPU with no GPU. Got ~40 tok/s single-stream (lossless) and ~124 batched. The surprising part was the byte budget: for this model you compress the output head (32% of per-token bytes), not the experts (16%). The writeup has the bandwidth roofline and the dead-ends; the repo has the reproducible recipe. Happy to answer questions. Repo: https://github.com/arun-prasath2005/gemma4-cpu-moe
More ai this month
the category →
Paintings by AI models: programs run through a simulation of oil paint.
AI · 22d ago · stillwet.art
FastRouter.ai▲364Route requests to the right LLM for cost, latency & quality
AI · 23d ago · fastrouter.ai
Source: Show HN post ↗
Launched alongside, April 2026
the whole month →
- AG
Thought the resources for GPU arch were lacking, so here we are
Life & fun · Apr 2026 · jaso1024.com
Built a ~9M param LLM from scratch to understand how they actually work. Vanilla transformer, 60K synthetic conversations, ~130 lines of PyTorch. Trains in 5 min on a free Colab T4. The fish thinks the meaning of life is food. Fork it and swap the personality for your own character.
AI · Apr 2026 · github.com
I made a laptop stand out of solid concrete, in the style of brutalist architecture, with themes of urbex and urban decay
Life & fun · Apr 2026 · sam-burns.com
With social media and now AI, its important to keep the indie web alive. There are many people who write frequently. Blogosphere tries to highlight them by fetching the recent posts from personal blogs across many categories. There are two versions: Minimal (HN-inspired, fast, static): https://text.blogosphere.app/ Non-minimal: https://blogosphere.app/ If you don't find your blog (or your favorite ones), please add them. I will review and approve it.
AI · Apr 2026 · text.blogosphere.app