Inspect the checks
Review the code, execution settings, results, and verification record. Checks that fail or could not run stay visible.
A self-evolving quantum solution platform.
Generate, run, and reuse quantum circuits.
Follow a circuit from generation and verification to editing in Studio and reuse through the Atlas.
How it fits together
From a question to a result you can check.
Leona is five parts on one thread. You ask, build, look things up, learn and share, and each part hands its work to the next.
01 · Nala
Ask in plain words
Describe what you want. Nala writes the circuit, runs it, and shows you the checks it passed, so you start from a working answer rather than a blank file.
Open Nala02 · Studio
Edit, run, keep versions
Every circuit opens in Studio. Change a gate, simulate it, verify it, and save the version. Your work stays private until you decide otherwise.
Open StudioA Bell state, step by step
from qiskit import QuantumCircuit
qc = QuantumCircuit(2)
qc.h(0)
qc.cx(0, 1)
qc.measure_all()Create a superposition with H, entangle the two qubits with a controlled X, then measure.
03 · Quantum Atlas
Look it up at the source
A public reference of algorithms, papers and circuits, each tied to the paper it comes from. Pull one into Studio or cite it.
Browse the Atlas04 · Notebooks
Learn by running it
Turn a question into a Jupyter lesson with cells that run and exercises that check themselves.
Open Notebooks05 · Qapps
Share it as a small app
Wrap a circuit in a few inputs and a chart, then send the link. Anyone can try it without touching code.
Open QappsCode and evidence
Keep the work behind each result
Continue your research
Save a circuit in Studio, edit its implementation, and build on the next version. The public Atlas provides reusable references.
Choose what to share
Your Studio work is private by default. Publishing is a separate action you control.
Work in familiar frameworks: Qiskit · Cirq · PennyLane
Benchmarks
Measured on code that has to run.
Benchmark results from our own evaluations and published sources. Methods and conditions differ; see the note below.
LeonaQ, our evaluationReported in the source
- LeonaQ56.3%
- Qiskit Code Assistant47.0%
- Granite 8B Code QK46.5%
- DeepSeek Coder 33B39.6%
- CodeLlama 34B Python26.7%
- LeonaQ72.3%
- Gemini59.5%
- GPT57.1%
- Claude45.2%
- LeonaQ71.4%
- Gemini54.8%
- GPT52.4%
- Claude35.7%
- LeonaQ66.7%
- GPT42.9%
- Gemini40.5%
- Claude26.2%
LeonaQ's Qiskit HumanEval figure comes from the dated run on the Benchmarks page, which names the model, code and dataset commits, date and spend. It also explains the 45 reference solutions blocked by our sandbox; generated alternatives can still pass those tasks. The three QuanBench+ figures are internal evaluations. Their model, prompts, task-level results and scoring method (including the denominator and retry budget) have not been published. Comparator figures are the ones reported in the Qiskit HumanEval paper, Qiskit's model card and the QuanBench+ paper; Gemini 3 Pro, GPT-5.1 and Claude 3.7 Sonnet appear under their family names. Datasets, prompts and environments differ between those runs and ours, so read this as directional rather than a controlled head-to-head.
SourcesQiskit HumanEval paper · Table II ↗Qiskit model card · benchmark table ↗IBM Quantum · Qiskit Code Assistant ↗QuanBench+ paper · Table 3 ↗
Start with a question or a circuit.
Describe your task to Nala, or find a starting point in the Atlas.