DeepCoder: An Fully Open-Source 14B Coder at O3-Mini Level

pretty-radio-b75.notion.site

15 points·sijuntan··6 comments

Through a joint collaboration between the Agentica team and Together AI, we release DeepCoder-14B-Preview, a code reasoning model finetuned from Deepseek-R1-Distilled-Qwen-14B via distributed RL. It achieves an impressive 60.6% Pass@1 accuracy on LiveCodeBench (+8% improvement), matching the performance of o3-mini-2025-01-031 (Low) and o1-2024-12-17 with just 14B parameters. We’ve open-sourced our dataset, code, training logs, and systems optimizations for everyone to progress on scaling and accelerating intelligence with RL.

What sort of hardware is needed to run a model like this at a vaguely useful speed?

Depends on what you consider vaguely useful, but any modern GPU which can fit the weights at a reasonable quantization (so, 16+ GB, maybe 12 in a pinch) will probably be tolerable. Bare minimum is probably ~3060 12GB.

Thanks! So this could actually be worth running on a mid-specced recentish Mac laptop?

Macs are a bit odd because you're using system memory (so, more but slower) and Apple's APIs, but it's definitely worth a try. I'd guess >= M2 Pro + >= 24 GB RAM would do okay. Also maybe try an MLX version? (e.g. https://huggingface.co/justinmeans/DeepCoder-14B-Preview-mlx...)

Sweet. Thank you so much!

Reply on news.ycombinator.com