iiinova — where your digital self keeps evolving

Juan Cruz-Benito @juancb.bsky.social · 16/09/2026

The rest of the days, I’ll be attending other talks and stopping by the booths, if you’re interested in our work at IBM Research and IBM Quantum or just want to say hi, don’t be shy!

000

Juan Cruz-Benito @juancb.bsky.social · 16/09/2026

Hello IEEE Quantum Week 2026 conf! Thrilled to attend this great conference again. For those who are around, I’ll be giving a talk tomorrow about our work in AI for Quantum Error Correction at the “AI for Circuit Synthesis, Optimization, and Discovery” workshop.

121

Juan Cruz-Benito @juancb.bsky.social · 29/05/2026

Models implement known algorithms well (Simon's 82%) but struggle to encode problems into quantum terms (SAT+Grover's 34%). Most failures are bad quantum reasoning, not syntax. And chain-of-thought helps reasoning-tuned models but hurts the rest. Paper arxiv.org/abs/2605.27210

arxiv.org

Qiskit QuantumKatas: Adapting Microsoft's Quantum Computing exercises for LLM evaluation

We adapt Microsoft's QuantumKatas -- a well-established quantum computing curriculum -- from Q# to Qiskit, the most widely-adopted quantum computing framework, and package it with an evaluation framew...

100

Juan Cruz-Benito @juancb.bsky.social · 29/05/2026

New preprint with @ismaelfaro.bsky.social : "Qiskit QuantumKatas — a benchmark for evaluating LLMs on quantum code". We translated Microsoft's QuantumKatas to Qiskit (350 tasks, 26 categories) and ran 16 models -frontiers, open weights- × 7 prompting setups = 39,200 runs.

110

Juan Cruz-Benito @juancb.bsky.social · 22/01/2026

The MCP server supports: - Permutation synthesis (SWAP routing) - Linear function synthesis (CNOT circuits) - Clifford synthesis (H, S, CNOT gates) This is how we envision training AI-powered transpilation models starting now. #QuantumComputing #AI #RL #Qiskit #MCP

100

Juan Cruz-Benito @juancb.bsky.social · 22/01/2026

The key difference: instead of fine-tuning LLMs, we're training RL agents that learn to synthesize quantum circuits optimized for real hardware.

100

Juan Cruz-Benito @juancb.bsky.social · 22/01/2026

Inspired by @hf.co's HF Skills Training (where Claude can submit fine-tuning jobs via natural language), we built something similar for quantum circuit synthesis with the new qiskit-gym-mcp-server proof of concept.

100

Juan Cruz-Benito @juancb.bsky.social · 22/01/2026

✨We got AI agents to train RL models for quantum transpilation! What if you could tell an AI assistant: "train a model to synthesize linear function circuits for X qubits on a ibm_nighthawk" - and it just does it?

110

Juan Cruz-Benito @juancb.bsky.social · 02/12/2025

With our scripts, you can: ✅ Download the LLM optimized for local enviroments (GGUF) ✅ Enable inference on your machine (through Ollama) ✅ Automatically configure the Qiskit Code Assistant VSCode or JupyterLab extension Give it a try and share your feedback—we’d love to hear how you use it!

110

Juan Cruz-Benito @juancb.bsky.social · 02/12/2025

Don’t have access to the Qiskit Code Assistant because you’re not on an IBM Quantum Premium plan? No worries—you can run it directly on your computer! We’ve recently improved the local setup experience.

Instructions on how to setup the Qiskit Code Assistant in local environments easily

100

Juan Cruz-Benito @juancb.bsky.social · 27/11/2025

✅ Improved UX: More precise responses for quantum programming tasks. ✅ Open access: Available on Hugging Face → huggingface.co/Qiskit/mistr... This update is part of our mission to make quantum development more intuitive and efficient. Explore it, test it, and share your feedback!

huggingface.co

Qiskit/mistral-small-3.2-24b-qiskit · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

120

Juan Cruz-Benito @juancb.bsky.social · 27/11/2025

We’ve upgraded the Qiskit Code Assistant! Last month, we introduced mistral-small-3.2-24b-qiskit, replacing granite-3.3-8b-qiskit. ✅ Better accuracy: The new model ranks higher across key benchmarks, including QiskitHumanEval, QiskitHumanEval Hard, HumanEval, MathQA, ASDiv, SciQ and others.

huggingface.co

Qiskit/mistral-small-3.2-24b-qiskit · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

100

Juan Cruz-Benito @juancb.bsky.social · 10/11/2025

🔗 Check out the full survey results: Unitary Fund Survey 2025 unitaryfoundation.github.io/survey-2025

unitaryfoundation.github.io

2025 Quantum Open Source Survey

Unitary Foundation

000

Juan Cruz-Benito @juancb.bsky.social · 10/11/2025

And thank you to the community for your continued trust and feedback. Let’s keep pushing the boundaries of quantum + AI software together. #QuantumComputing #Qiskit #IBMQuantum #AITranspiler #UnitaryFund #QuantumSoftware #OpenSource #AIforQuantum

120

Juan Cruz-Benito @juancb.bsky.social · 10/11/2025

Huge kudos to the amazing team behind this David Kremer, Víctor Villar Pascual, Jesús Talavera Gómez, Yaiza García Martín-Mantero, @ismaelfaro.bsky.social — your dedication, creativity, and deep technical insight are what make this possible.

100

Juan Cruz-Benito @juancb.bsky.social · 10/11/2025

This makes advanced AI-powered transpilation more accessible than ever. The Unitary Fund survey is widely recognized as the primary pulse check for quantum software adoption and developer preferences. Seeing our work reflected there is both humbling and energizing.

100

Juan Cruz-Benito @juancb.bsky.social · 10/11/2025

One of the most exciting updates this year: our AI transpiler passes — including the latest Pauli Networks pass (arXiv:2503.14448) — can now run in local mode, removing the need for an IBM Quantum premium plan.

100

Juan Cruz-Benito @juancb.bsky.social · 10/11/2025

In less than two years, the package has surpassed 929K downloads, becoming a relevant tool for quantum circuit optimization/transpilation across the community.

100

Juan Cruz-Benito @juancb.bsky.social · 10/11/2025

🚀 I'm thrilled to share that the qiskit-ibm-transpiler library has moved up to #3 in the @unitary.foundation 2025 Survey under Full-stack development platforms, compilers, and simulators. Last year, we were #4 — and this year’s leap is a testament to the incredible work of the team behind it.

120

Juan Cruz-Benito @juancb.bsky.social · 08/07/2025

We'd love to hear how you're making the most of these models! Share your experiences and quantum development journey with us. 🌟 Big thanks to the team involved in this release! cc/ Nicolas Dupuis, Adarsh Tiwari, @ismaelfaro.bsky.social #QuantumComputing #OpenSource #AI #LLMs #Qiskit #IBM

031

Juan Cruz-Benito @juancb.bsky.social · 08/07/2025

🔗 Browse All Models: huggingface.co/Qiskit 🔗 Get Started: Learn how to use these models locally in our comprehensive documentation: quantum.cloud.ibm.com/docs/en/guid...

huggingface.co

Qiskit (Qiskit)

Creating a link between Quantum Computing and AI

100

Juan Cruz-Benito @juancb.bsky.social · 08/07/2025

Available Models: • Granite 3.3 8B Qiskit (+ GGUF) • Granite 3.2 8B Qiskit (+ GGUF) • Qwen2.5-Coder 14B Qiskit (+ GGUF) All models are available on Hugging Face and ready for local deployment!

100

Juan Cruz-Benito @juancb.bsky.social · 08/07/2025

🔬 Expanding Our Horizons - We continue to leverage the newest stable versions of IBM Granite 3.2 and 3.3 models, while for the first time exploring beyond IBM Granite with the powerful Qwen2.5-Coder series

100

Juan Cruz-Benito @juancb.bsky.social · 08/07/2025

✨ Qiskit 2.0 Compatibility - Our newer models are fully compatible with Qiskit 2.0, ensuring seamless development with the latest version 🧠 Enhanced Text Understanding - Significant improvements in general text comprehension and code generation capabilities

100

Juan Cruz-Benito @juancb.bsky.social · 08/07/2025

🚀 We're thrilled to announce the latest open-source LLM releases from the Qiskit Code Assistant team! 🚀 These new models bring enhanced capabilities and broader compatibility to quantum computing development. What's New: 🧵👇

110

Juan Cruz-Benito @juancb.bsky.social · 07/06/2025

Yesterday I returned to my alma mater, @usaloficial.bsky.social, to share my PhD journey with current doctoral students. Encouraged them to make the most of this adventure in preparation for what’s next! Thanks for the invitation, it’s always a pleasure 🎓 #PhD #mentorship #research

021

Juan Cruz-Benito @juancb.bsky.social · 28/03/2025

Just returned from the exclusive IBM Tech 2025 in Singapore! 🇸🇬 It's been incredible to connect with IBM's brightest technical minds exploring the future of technology. The most powerful innovations happen when diverse expertise collides! Grateful and inspired. #IBMTech2025

120

Juan Cruz-Benito @juancb.bsky.social · 19/03/2025

In case you want to play with this new AI transpiler pass, you don't need to wait for a release. This AI pass has been available in the Qiskit Transpiler Service since last November 2024, as presented in the Quantum Developer Conference 2024.

110

Juan Cruz-Benito @juancb.bsky.social · 19/03/2025

David Kremer will be presenting this work at the APS March Meeting in Anaheim on Wednesday (March 19) morning - do check it out if you are at the conference! summit.aps.org/events/MAR-L... If you're around, check it out

summit.aps.org

Quantum Circuits and Complexity

8:00 am – 11:00 am, Wednesday March 19, Session MAR-L35, Anaheim Convention Center, 256B (Level 2)

100

Juan Cruz-Benito @juancb.bsky.social · 19/03/2025

⚠Paper alert! In this new paper we introduce a RL-based synthesis pass for Pauli Networks (Clifford + arbitrary angle Pauli rotation circuits), and an optimization pass for general circuits that uses this Pauli Network synthesis pass. arxiv.org/abs/2503.14448

arxiv.org

Pauli Network Circuit Synthesis with Reinforcement Learning

We introduce a Reinforcement Learning (RL)-based method for re-synthesis of quantum circuits containing arbitrary Pauli rotations alongside Clifford operations. By collapsing each sub-block to a compa...

110

Juan Cruz-Benito @juancb.bsky.social · 14/03/2025

But that's not all - our update also includes significant improvements to the benchmark, making it more robust and rigorous in terms of code execution tests to provide even more accurate and comprehensive evaluations of your LLM-generated quantum code

110

Juan Cruz-Benito @juancb.bsky.social · 14/03/2025

New Qiskit HumanEval release! 🚀 We've just updated huggingface.co/datasets/Qis... to be compatible with the latest Qiskit 1.4 release!

110

Juan Cruz-Benito @juancb.bsky.social · 10/03/2025

Exciting update! The Qiskit Code Assistant service now exposes compatible endpoints with OpenAI's Completions API. This integration enables seamless usage via existing libraries (OpenAI, LiteLLM, etc.), making it easy to infuse Qiskit knowledge into your LLM pipelines.

The image depicts how to use the Qiskit Code Assistant through the OpenAI completions-compatible available endpoints using Python a and LiteLLM

100

Juan Cruz-Benito @juancb.bsky.social · 05/03/2025

Oh! I was assuming you were getting outdated code from our model. Despite it may happen, we have measured that most of the times it should give you the right answer (even more when comparing it vs other AI models that are not specialized on Qiskit as you comment)

120

Juan Cruz-Benito @juancb.bsky.social · 05/03/2025

Thanks for the interest @quantumyakar.bsky.social. We thank the feedback from our users We'd expect that, the new models are better in our internal benchmarks about outdated code. In any case, this model is a release candidate and needs more testing before promoting it to be a stable release.

110

Juan Cruz-Benito @juancb.bsky.social · 05/03/2025

🙌 Thanks to the feedback received, we've released a revised version of the granite-8b-qiskit-rc-0.10! Check out the improved checkpoint at huggingface.co/Qiskit/grani...

huggingface.co

Qiskit/granite-8b-qiskit-rc-0.10 · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

240

Juan Cruz-Benito @juancb.bsky.social · 18/02/2025

Trained on a significantly expanded Qiskit synthetic dataset, this release marks the end of an era - it's our final model using the current training approach. What's next? We're pivoting to newer Granite base models and starting to integrate other cutting-edge techniques. Stay tuned!

110

Juan Cruz-Benito @juancb.bsky.social · 18/02/2025

🚀 Last Friday, we shipped something special: granite-8b-qiskit-rc-0.10, our latest revision of the LLMs that empower the Qiskit Code Assistant.

130

Reposted by Juan Cruz-Benito

Steven Heidel @stevenheidel.com · 26/01/2025

twitter discourse would be improved by adding chains of thought before posting

1212