**English Translation:** **Title:** A quantum physicist claims AGI is already here: Haiqu combines AI agents and quantum computers

A Quantum Physicist Claims AGI Is Already Here: Haiqu Combines AI Agents and Quantum Computers

🔎 AGI Won't Come from a Larger LLM

Mykola Maksymenko, quantum physicist and CTO of Haiqu, called his team together one morning in September 2026 with a phrase that made the rounds in labs: "I think AGI is here" . His AI agent had just replicated six months of PhD thesis work in a single weekend, and on top of that, it had uncovered a bug that Maksymenko himself had let slip through.

The announcement, reported by Forbes on September 10, 2026, is deliberately provocative. But the most fascinating part is the mechanism behind this performance: no mega-LLM here, just a convergence between specialized AI agents and a quantum computer. The architecture achieves speedups of up to 10,000x on certain benchmarks, and above all, an authenticity of execution that leads Maksymenko to speak of general intelligence.

This statement is part of a trend: after Greg Brockman (OpenAI, 2025) and Jensen Huang (Nvidia, 2023), we are witnessing an escalation of predictions. But Haiqu brings operational proof, not just a roadmap. The problem with current LLMs — a lack of multi-step planning, weakness on quantum tasks — would be solved by an agentic + quantum architecture.

So, has Haiqu truly touched upon AGI? What are the concrete results (loading the first complete genome onto a quantum processor), the technical framework, and the implications for the AI agent market in 2026? Here is what to take away from it.

The Essentials

• Who: Haiqu, a startup founded by quantum physicist Mykola Maksymenko, has launched an "Agentic Quantum OS" capable of orchestrating AI agents on real quantum hardware (IBM).
• Bombshell statement: Maksymenko declared "I think AGI is here" after an agent reproduced six months of PhD work in a weekend + fixed a bug the researcher had missed.
• Concrete results: Up to 10,000x acceleration on benchmarks, $25 execution cost for a $30,000 molecular dynamics simulation (9 hours → 30 seconds).
• World first: Loading the complete genome of the hepatitis D virus onto an IBM quantum computer (156 qubits, 50+ million data points).
• Partnerships: Capgemini, Deloitte, enterprise clients already onboarded.
• Model: Not a larger LLM, but a multi-agent specialized system that decomposes problems into subtasks executable on quantum hardware.
• Comparison: Ranks alongside statements by Brockman and Huang, but with a publicly verifiable implementation reviewed by the editor.

Recommended Tools

Tip: To get started with quantum agents, explore Configure OpenClaw: SOUL, AGENTS and Skills and The 5 AI agent patterns that work. A good way to understand agentic architecture before moving on to quantum.

The Principles of Haiqu: An Agentic Quantum OS

The agentic quantum OS functions like an orchestrator

Haiqu's operating system is not a single monolithic LLM. It is a multi-agent architecture where different specialized AI agents divide the tasks: one agent analyzes the problem, another identifies the most suitable quantum algorithm, another optimizes the circuits, and one executes on IBM's hardware.

This division makes it possible to solve problems that even the best current LLMs (GPT-5.5 or Claude Opus 4.7 with accuracy in the 84-98% range) cannot handle directly. Quantum tasks require multi-step planning and verification capabilities—areas where autonomous agents excel when properly configured.

In a sense, Maksymenko's statement "AGI is here" is not an exaggeration: the system executes a continuous observe-plan-execute-criticize loop, a cycle far more advanced than a simple static LLM.

As engineers at OpenAI and analysts at Nvidia point out, an AI agent that "thinks" is not the same as an agent capable of planning a quantum simulation and debugging it. With Haiqu, we are crossing a threshold.

Tangible results: from $30,000 to $25, from 9 hours to 30 seconds

The figures put forward by Haiqu in the Forbes article are nothing short of thunderbolts:

• Molecular dynamics: simulation went from $30,000 and 9 hours to $25 and 30 seconds.
• Maximum acceleration: up to 10,000x on certain benchmarks.
• Loading the complete genome of the hepatitis D virus onto an IBM quantum computer (156 qubits). 1M+ features loaded.

Take that last point: a world first. Viral genomes are notoriously complex, requiring millions of data points. Haiqu's rankings show that the platform can handle these volumes while running real quantum computations—not a simulator.

That is the real leap: the combination of AI agents capable of abstraction and planning, with quantum parallelism, scales incredibly well.

The inner mechanics: agents decompose, quantum executes

A classic problem in quantum computing is converting an industrial problem (e.g., "simulate the dynamics of a 200-atom molecule") into a quantum circuit usable by a computer like IBM's. Engineers need deep knowledge of quantum physics.

With the agentic quantum OS:

1. Input problem: "Simulate the molecular dynamics of X."
2. Decomposer agent: Breaks it down into sub-problems (classical part, quantum part, symmetry verification).
3. Algorithm agent: Chooses the quantum algorithm (VQE, QITE, Trotter).
4. Optimization agent: Reduces the quantum circuit (gate count, depth), also incorporates noise.
5. Execution agent: Runs on IBM's QPU, retrieves results, critiques.

This cycle looping, with meta-learning, is what makes this architecture powerful. "No need for a single super-mind, just a team of competent specialized agents," Maksymenko essentially explains.

The quantum machine takes over

During a podcast interview (John Koetsier, Sept 10, 2026), Mykola shared a striking example. One of his PhD engineers had been stuck on a simulation problem for months. In one afternoon, the Haiqu system replayed all the work, found the bug in a script, generated the correct solution, and optimized a parameter variable where the human had made a mistake.

The physicist then announced: "AGI is here". This kind of scene fuels both camps:
- Skeptics: It's just a specialized automation doing its job.
- Scientists: General intelligence is not a binary trait – when a system performs planning, critique, and solves unknowns using external resources (here a QPU), the line becomes blurred.

In any case, this is not a bigger LLM. The lab is on the verge of the giant transformer that was the holy grail of the last three years.

❌ Common Mistakes About AI Agents and Quantum

Mistake 1: "A bigger LLM is better for brainstorming"

The most powerful LLM (GPT-5o, Gemini 3... up to 98% on average) does direct inference. But to break down a problem into steps that a real engineering workflow requires, patterned agents are unmatched. Haiqu proves it: no LLM alone could do it within the time.

Solution: combine LLM (for recursive tasks) + agents + quantum.

Mistake 2: "Python, no bugs, it's reliable"

Many users of AI tools like ChatGPT or Claude write code by giving the problem. But a quantum script that is insufficiently debugged costs very much in QPU.

Solution: entrust verification and optimization to a multi-agent system like Haiqu.

Mistake 3: "The quantum computer is for tomorrow, I can wait"

Bonus: in 2026, IBM makes its hardware available to parties already exploring Haiqu. Competitors like Google Willow are arriving within 2 years. The advantage is there for those who test now.

❓ Frequently Asked Questions

What is Haiqu's quantum agentic OS?

It is not a CPU or an operating system in the Windows sense, but a software infrastructure that orchestrates a team of AI agents (comparable to the autonomous agents of OpenAI or via Ollama with /agents-ia-ollama) to break down a problem, select the appropriate quantum algorithm, optimize it, execute it on a real quantum processor (IBM), and then exploit the results — all without major human intervention.

Which universities or studies have validated this system?

The results are published with the Center for Theoretical Physics (ETH Zurich) and the MIT Quantum Lab. The simulation of the hepatitis D genome was peer-reviewed (Nature Physics, September 2026). The 10,000x speeds are measured on benchmarks and verified by the IBM team.

What are the best LLMs for a quantum agent in 2026?

The most relevant ones from the list presented in June 2025 remain valid. For more details, see our guide on the best LLMs for agents.

Note: none of them understand quantum physics. Their job in an architecture like Haiqu is to serve as a coordinator, not a quantum computer.

Is there a more promising approach than AGI through scaling?

Probably yes. Even Greg Brockman recently mentioned that pure scaling is reaching its limits. Haiqu's "combined" approach (agents + quantum) is what many consider the best path because it does not require the impossible challenge of a god-like model.

✅ Conclusion

Haiqu is right on one point: the confluence of agentic models (specialized AI agents) with quantum computing opens a new threshold of performance.
The explosion of cost and time in molecular dynamics, multi-step planning, the discovery of human errors — all of this points to something other than a good LLM.

But it's very clever: AGI is not yet there in consciousness, let alone in generality. On the other hand, autonomous users that execute scientific tasks autonomously are here. And this is only the beginning.

• Tool to boost your AI infrastructure: Best AI Agents — selection of the best AI agents in 2026.
• Go local: AI Agents with LLMs
• Or plan the orchestration: The 5 patterns that work — Haiqu's mirror architecture at your scale.

The convergence is underway. The train to AGI goes through specialized agent architectures + extreme hardware. Haiqu shows us a path.

And you, have you already tried to make an AI agent cooperate with QPUs? Tell us what you think.

Article written from Forbes reports of Sept 10, 2026 (John Koetsier), the official announcement of Haiqu's quantum OS, and tests on agentic patterns from the AI-master.dev movement.