Aurora-Coder-Base

---

1. Introduction

Aurora-Coder-Base is a code-specialised checkpoint fine-tuned on a deduplicated multi-language corpus. It targets repository-level code completion and agentic tool use.

2. Evaluation Results

Comprehensive Benchmark Results

	Benchmark	StarBase	CodeNova	Aurora-mini	Aurora-Coder-Base
Core Reasoning Tasks	Math Reasoning	0.579	0.535	0.552	0.592
	Logical Reasoning	0.810	0.831	0.824	0.846
	Common Sense	0.731	0.734	0.750	0.761
Language Understanding	Reading Comprehension	0.681	0.711	0.673	0.732
	Question Answering	0.607	0.592	0.591	0.628
	Text Classification	0.833	0.808	0.832	0.843
	Sentiment Analysis	0.785	0.780	0.756	0.806
Generation Tasks	Code Generation	0.676	0.665	0.642	0.692
	Creative Writing	0.609	0.623	0.639	0.656
	Dialogue Generation	0.657	0.616	0.618	0.673
	Summarization	0.762	0.780	0.748	0.787
Specialized Capabilities	Translation	0.808	0.778	0.808	0.816
	Knowledge Retrieval	0.655	0.645	0.691	0.697
	Instruction Following	0.725	0.743	0.742	0.779
	Safety Evaluation	0.715	0.700	0.703	0.759

Overall Performance Summary

The Aurora-Coder-Base demonstrates strong performance across all evaluated benchmark categories, with particularly notable results in reasoning and generation tasks.

3. Chat Website & API Platform

We offer a chat interface and API for you to interact with Aurora-Coder-Base. Please check our official website for more details.

4. How to Run Locally

Please refer to our code repository for more information about running Aurora-Coder-Base locally.

Temperature

We recommend setting the temperature parameter to 0.6.

5. License

This repository is released under the mit license. The model supports commercial use.

6. Contact

If you have any questions, please contact us at hello@aurora.dev.