Wayve launches GAIA-3, advancing world models from simulation to evaluation

LONDON, December 2, 2025 – Wayve, the leader in Embodied AI, today announced the launch of GAIA-3, a new generation of its generative world model designed to accelerate the evaluation and validation of its autonomous driving AI. Building on the success of GAIA-2, this new model is larger, more capable, and purpose-built to enable meaningful evaluation of modern end-to-end driving systems.

Real-world safety hazards are rare and unsafe to reproduce, and current testing relies heavily on controlled test-track experiments that cannot capture the complexity of real-world environments. Generative world models address this challenge by simulating dynamic driving scenarios that allow AI models to learn, plan, and make decisions safely and efficiently.

At 15 billion parameters, GAIA-3 is double the size of GAIA-2, with a video tokenizer that is also twice as large, enabling a more faithful representation of real-world physics and causality. It is pre-trained on ten times more data, spanning multiple continents, vehicle types, environments, and driving conditions. The model produces higher-fidelity video generations, with sharper visuals, more consistent lighting, and richer texture detail, particularly in road signage.

GAIA-3 also introduces new evaluation modes, including:

• Safety-critical scenario generation to test “what-if” edge cases safely offline
• Embodiment transfer for consistent evaluation across different vehicle rigs
• Controlled visual diversity for robustness testing under changing conditions

GAIA-3 advances world modeling from visual synthesis to autonomy evaluation. It can generate structured, realistic scenes that support quantitative assessment of AI-driving behavior under many different conditions. Early studies show that GAIA-3 simulated testing closely mirrors real-world driving results and has reduced synthetic-test rejection rates fivefold.

Jamie Shotton, Chief Scientist at Wayve, said:
“With GAIA-3, Wayve is taking a bold step forward, advancing world modeling from visual synthesis to true autonomy evaluation and validation. The model learns to recreate the dynamics of real-world environments – from everyday traffic to rare, safety-critical events. It will enable developers to measure, compare, and accelerate progress toward safe and scalable autonomous driving.”

Wayve is also working in partnership with Warwick Manufacturing Group at the University of Warwick on DriveSafeSim, a UK government-funded project, to validate the use of generative world models such as GAIA-3 in safety evaluation.

For more details on GAIA-3’s architecture, methodology, and evaluation, please refer to our technical blog here.