Qiskit Code Assistant
A family of LLMs (8B–24B parameters) specialized for quantum code generation. Models built on IBM Granite, Mistral, and Qwen foundations, fine-tuned on curated Qiskit datasets including Python scripts, Jupyter notebooks, and synthetic Q&A pairs.
Open-source model family:
- Mistral Small 24B — Largest model, highest accuracy
- Qwen2.5 Coder 14B — Strong coding foundation
- Granite 3.x 8B — Efficient, multiple versions
- GGUF quantized versions available for local deployment
Achieves 46.53% on Qiskit HumanEval—significantly outperforming competing models (24.75%–39.6%). Supports natural language to code (“define a Bell circuit and run it on ibm_brisbane”) and intelligent autocomplete.
We created and open-sourced Qiskit HumanEval and Qiskit HumanEval Hard—benchmarks with 150+ tasks each, now used industry-wide for evaluating quantum code LLMs. Integrated into VS Code and JupyterLab.
Posts
Run Qiskit Code Assistant Locally easily!
You can now run the Qiskit Code Assistant locally easily. Download optimized models in GGUF format, install Ollama, and configure your VSCode or JupyterLab extension with a single command.
Qiskit Code Assistant Upgraded to Mistral-Small-3.2-24B-Qiskit
We’ve upgraded the Qiskit Code Assistant! Last month, we introduced mistral-small-3.2-24b-qiskit, replacing granite-3.3-8b-qiskit, delivering better accuracy across key benchmarks and more precise responses for quantum programming tasks.
Qiskit Code Assistant: New Open-Source LLM Models Released
Announcing the latest open-source LLM releases from the Qiskit Code Assistant team, featuring Qiskit 2.0 compatibility, enhanced text understanding, and new models including Granite 3.3, Granite 3.2, and Qwen2.5-Coder series.
New Qiskit HumanEval Release: Qiskit 1.4 Compatibility and Benchmark Improvements
Released a new version of Qiskit HumanEval compatible with Qiskit 1.4, featuring significant improvements to the benchmark including more robust and rigorous code execution tests for more accurate evaluations of LLM-generated quantum code.
Qiskit Code Assistant Now Compatible with OpenAI Completions API
Exciting update: Qiskit Code Assistant service now exposes compatible endpoints with OpenAI Completions API. This enables seamless usage via existing libraries like OpenAI and LiteLLM, making it easy to infuse Qiskit knowledge into your LLM pipelines.
Granite-8B-Qiskit-RC-0.10: Updated Checkpoint Using Current Training Approach
Released granite-8b-qiskit-rc-0.10, the latest revision of the LLMs that empower Qiskit Code Assistant. Trained on significantly expanded Qiskit synthetic dataset, this marks the final model using the current training approach as we pivot to newer Granite base models and cutting-edge techniques.
Shipping Granite-8B-Qiskit-RC-0.10: The Final Model of Current Training Era
Shipped granite-8b-qiskit-rc-0.10, the latest revision of LLMs empowering Qiskit Code Assistant. Trained on significantly expanded Qiskit synthetic dataset, marking the end of an era as the final model using current training approach before pivoting to newer Granite base models.
Introducing Qiskit Code Assistant: New Blog Post Published
Published a comprehensive blog post summarizing what Qiskit Code Assistant is and how to start using it. Following the recent launch, users are actively leveraging the tool features, and the team is developing improved models with enhanced capabilities for open source release.
Presenting Qiskit HumanEval at IEEE Quantum Week 2024
Presenting the Qiskit HumanEval benchmark for LLMs at IEEE Quantum Week 2024 in the SYS-BNCH Benchmarking session. Available afterwards at the IBM Quantum booth to discuss AI and quantum computing initiatives.
Qiskit HumanEval: Evaluation Benchmark for Quantum Code Generation Published
Published research paper introducing Qiskit HumanEval dataset for evaluating Large Language Models capability to generate quantum computing code. The dataset comprises more than 100 quantum computing tasks with prompts, solutions, test cases, and difficulty ratings, establishing benchmarks for generative AI tools in quantum code development.
Qiskit Code Assistant: Training LLMs for Quantum Code Generation Paper Published
Published research paper on training specialized LLMs for quantum computing code generation using Qiskit. Addresses unique challenges in quantum programming, including scarcity of quantum code examples and rapid field evolution. Our model outperforms existing state-of-the-art quantum computing models.
Publications
Quantum source code generation based on a modeling system
An approach for assisting in the generation of quantum source code. The approach may include receiving a quantum source code input with a …
Quantum Verifiable Rewards for Post-Training Qiskit Code Assistant
Qiskit is an open-source quantum computing framework that allows users to design, simulate, and run quantum circuits on real quantum …
Qiskit HumanEval: An Evaluation Benchmark For Quantum Code Generative Models
Quantum programs are typically developed using quantum Software Development Kits (SDKs). The rapid advancement of quantum computing …
Qiskit Code Assistant: Training LLMs for generating Quantum Computing Code
Code Large Language Models (Code LLMs) have emerged as powerful tools, revolutionizing the software development landscape by automating the …
Quantum and non-quantum source code translation
One or more systems, devices, computer program products and/or computer-implemented methods of use provided herein relate to a process to …