Quantum computers have a persistent headache for engineers: calibration. After a quantum processing unit (QPU) runs for a while, the coherence between qubits drifts and requires recalibration. Manual calibration can take anywhere from a few days to a week, even with an experienced team.
On April 14, NVIDIA unveiled a new solution to this problem: NVIDIA Ising, the world's first open-source family of quantum AI models, released under the Apache 2.0 license for free use.
Two Models, Two Problems
Ising is not a single model but two complementary pieces:
Ising Calibration (35B parameters)
This is a vision-language model specifically designed to "read" measurement results from quantum processors. Qubit state data is difficult to represent with ordinary data, but a combined image-plus-text representation works surprisingly well.
- Model input: various measurement data from qubits (supports superconducting, quantum dot, ion, neutral atom, and other platforms)
- Model output: calibration operation suggestions and automated execution
- Effect: reduces calibration time from days to hours
On the newly introduced QCalEval benchmark, Ising Calibration achieved:
- 3.27% higher than Gemini 3.1 Pro
- 9.68% higher than Claude Opus 4.6
- 14.5% higher than GPT-5.4
An open-source model outperforming three of the strongest closed-source flagships in this specialized domain.
Ising Decoding (two 3D CNNs)
Quantum computers are inherently error-prone, and quantum error correction (QEC) is the mechanism for detecting and correcting these errors in real time. Ising Decoding offers two versions: fast and accurate.
| Version | Parameters | Speed Improvement | Accuracy Improvement |
|---|---|---|---|
| Fast | ~912K | 2.5x | 1.11x |
| Accurate | ~1.79M | 2.25x | 1.53x |
The baseline is pyMatching, the current industry-standard decoder. Outperforming it in both speed and accuracy is a significant achievement.
In tests with 13 qubits and a 0.3% error rate, each decoding round took only 2.33 microseconds (FP16 precision).
Why NVIDIA Is Doing This
The quantum computer market is still small and not NVIDIA's core business, but several strategic logics are at play:
GPUs are the best classical computing companion for quantum computers. Quantum error correction, quantum circuit simulation, and quantum algorithm optimization all run on classical hardware. NVIDIA's CUDA-Q platform is a hybrid computing framework for GPU+QPU, and Ising is directly integrated into it. Even with an open-source model, the ecosystem remains NVIDIA's.
Early positioning in the fault-tolerant quantum computing (FTQC) race. For quantum computers to become truly practical, fault tolerance is an unavoidable hurdle. Decoders are the most fundamental building block. Open-sourcing it now creates first-mover advantage when the market matures — much like what CUDA did for GPU computing.
Expanding the academic partner network. Harvard School of Engineering, Fermi National Accelerator Laboratory, Lawrence Berkeley National Laboratory, the UK's National Physical Laboratory, and IQM Quantum Computers — these institutions are core nodes in quantum computing research. Ising embeds NVIDIA into this supply chain.
Stock Market Ripple Effects
CNBC reported on April 16 that quantum computing stocks collectively rose after the Ising launch, with IonQ surging over 20% in a single day.
The rally signals that the market read the message: IonQ is one of NVIDIA's partners in quantum processors, and Ising supports IonQ's trapped-ion quantum hardware. In other words, NVIDIA has publicly endorsed the trapped-ion technical path — at least in the two critical capabilities of calibration and decoding.
The "quantum + AI" combination now has concrete products, not just slideware.
A Detail Worth Noting
The name Ising comes from the Ising model in physics, a statistical mechanics model used to describe phase transitions and magnetism. The calibration problem in quantum computers has a mathematical correspondence to this model, so the name is not arbitrary.
NVIDIA has given this technical direction a serious physics name, not a marketing gimmick.
No one can give a precise timeline for when quantum computers will become commercially viable. But if that day comes, quantum error correction is an infrastructure problem that must be solved. NVIDIA has planted a stake here — and open-sourced it.
Sources: NVIDIA Launches Ising, the World's First Open AI Models to Accelerate the Path to Useful Quantum Computers (NVIDIA Newsroom); NVIDIA Ising Introduces AI-Powered Workflows to Build Fault-Tolerant Quantum Systems (NVIDIA Technical Blog); CocoLoop; Quantum stocks on pace for a massive week after Nvidia debuts AI models to boost the tech (CNBC)