binary-husky/gpt_academic
A Swiss Army knife for LLM-powered academic grunt work
This Chinese-born project wraps a web GUI around dozens of plugins that translate, polish, and dissect papers and code using whatever model you have handy.
Velocity · 7d
+60
★ / day
Trend
→steady
star history
What it does GPT Academic is a Python web interface that plugs into OpenAI, Claude, Qwen, GLM, DeepSeek, and roughly a dozen other LLMs. It ships with a kitchen-sink collection of plugins: arXiv paper translation, LaTeX proofreading, PDF full-text translation, code-base analysis for Python/C++/Java, and even a voice-input listener. Everything is driven through a browser UI with customizable hotkeys and a dark-mode toggle.
The interesting bit The project treats LLMs as interchangeable utilities rather than a single vendor lock-in. You can queue the same task against GPT-4, ChatGLM, and a local Llama2 instance in parallel, or hot-swap API keys mid-session by pasting a new one into the chat box. The “self-analysis” plugin is a neat dogfooding trick: it feeds the project’s own source code back to the model to generate an annotated architecture report.
Key highlights
- Modular plugin system with hot-reload; buttons are dynamically generated from
functional.pyso you can add custom actions without touching the core. - First-class support for Chinese models (Qwen, GLM4, Spark, Ernie) alongside the usual OpenAI/Claude suspects.
- arXiv and PDF pipelines extract, translate, and re-render papers while preserving formulas in both TeX and rendered form.
- Docker images bundle LaTeX dependencies for the full paper-processing workflow.
- A companion
void-terminalpip package (experimental) exposes all plugins as plain Python function calls, headless-GUI-free.
Caveats
- Documentation and default UI are Chinese-first; English speakers will need to hunt for the translated READMEs.
- The README warns that dependency versions in
requirements.txtare pinned for a reason—upgrading packages blindly may break things. - A new GUI frontend is “Coming Soon” (as of January 2026), so the current Gradio-based interface may be in flux.
Verdict Researchers and students who churn through papers in Chinese or English will get the most mileage. If you already have a favorite chat client and only need occasional summarization, this is probably overkill.