VENETA
The Enterprise World Model.
VENETA formulates a company's hardest operating decisions, runs the combinatorial part on IBM quantum systems, and returns the answer to the operator's own screen. The operating data never leaves the network.
Enterprise AI that stays inside the network.
VENETA reads a company's own operating systems in a read-only sandbox, simulates before it recommends, attaches the evidence, and records every step for governance and audit.
IBM Quantum
IBM Quantum, delivered.
VENETA delivers IBM Quantum to enterprises: we formulate the combinatorial decisions of telecom, retail, the public sector, finance, manufacturing and defense. Written in Qiskit, verified against the classical optimum, and integrated into operations through WorldModel. Telecom and retail instances have run on IBM quantum systems. Every result carries its backend, qubits, shots and gap.
VENETA Korea Inc., which contracts IBM Quantum delivery for VENETA, signed the IBM Quantum GSI (Global Systems Integrator) agreement on 30 September 2026.
Products
One engine, delivered four ways.
IBM Quantum delivery
Your operations question becomes an optimisation instance, runs on IBM quantum systems, is verified against the classical optimum, and comes back into your systems. One edition per industry VENETA serves.
VENETA Orchestrate
Backup and recovery automated end to end from one console.
VENETA WorldModel
The engine under every delivery. A world model is an AI system for mission-critical environments — it uses your own model, wraps it in a memory and self-improvement layer, decides from measured readings, and learns only what a person approves. No data leaves your network.
VENETA Sentinel
Every quantum job and every outbound SaaS call leaves the building shielded, scanned and written to an egress ledger.
project veneta
VENETA's research and engineering site: open benchmarks, measurement notes and whitepapers, every number with its conditions.
All products and services
How the four fit together, and which product answers which question.
How it works
Simulate before you act. Record everything.
An operational question is answered in one loop. VENETA reads the live systems through tools, forecasts with predictive AI, solves the combinatorial part on IBM Quantum, simulates the options, gates each one on the service-level band, and drafts a recommendation with its evidence. A person approves; the change goes through the operator's own system; the outcome is remembered for the next question.
Every step of that loop is traced, and a critic that fails closed checks the answer before a person sees it. That is what makes the loop governable and auditable.
VENETA WorldModelWhitepaper · October 2026
The First Condition for a Local LLM to Evolve into a World Model
Why the language models you run inside your own network fall short, and what changes when you put a memory and self-improvement layer around the model instead of replacing it. Twelve open models, 12B to about 400B, and eleven smaller ones, measured inside one network.
project veneta
The numbers live where you can check them.
VENETA builds and delivers the software. project veneta is where the same engineering group publishes what it measured: the benchmarks — veneta-memory-bench and veneta-worldmodel-bench — the measurement notes, and the whitepapers. Every number carries the conditions it was taken under, and what we have not measured is written down as plainly as what we have.
The second site is deliberate. A supplier's own numbers are worth more when the method sits apart from the page that sells the product: dated, versioned, and written so a customer, a partner or a reviewer can run it again and disagree with it. The veneta-memory-bench cases, grading code and protocol are published under Apache-2.0; the engine is not.
Industries
One engine, an edition for each industry.
Telecom
Fault correlation, remediation simulation, network-operations policy, energy saving.
Energy
Asset operation, load balancing, outage response and energy-saving schedules.
Semiconductor
Fab scheduling, tool allocation and data movement between fab systems.
Retail
Demand, assortment and store operations over the retailer's own systems.
Manufacturing
Process and equipment scheduling, material routing, inspection resources.
Defense
Readiness, logistics and asset-allocation questions in closed networks.
Quantum partnership
Quantum work for enterprises and public bodies starts with a conversation.
We take partnership enquiries, visit requests and materials requests. In a pilot workshop we decide together whether quantum earns its place on your problem.
News
Latest from VENETA.
94.5 % on LongMemEval, a benchmark we did not write
We put VENETA WorldModel's memory and self-improvement layer through a general-purpose memory benchmark we had no hand in. LongMemEval-S, all 500 questions, Claude Opus 5.5 as the writer, the benchmark's own judge, three runs per arm. With the layer on, 94.5 %; with it off, 19.7 %. Both arms get the same budget.
We measured the self-improvement learning curve
VENETA WorldModel's memory and self-improvement layer turns an operator's correction into a rule and does not apply it until a person approves. On six synthetic site conventions we measured whether that approval moves the outcome. With every proposal approved the score climbs 2 to 4 to 5.3 of 6 at sessions 2, 3 and 4, while the arm where nobody approved turned the same corrections into pending rules and stayed at 0.
IBM, AMD and Basque Quantum improve a quantum result with classical tensor networks
A preprint from IBM with AMD and Basque Quantum introduces an error-mitigation technique in which classical tensor networks correct the expectation values a quantum computer returns, run on 65 qubits of ibm_basquecountry, with the classical half up to 5.6 times faster on a GPU. Here is what the paper reports, and why a result that needs both machines is the shape we build for.