Google Launches Gemini 3 to Enable Single-Prompt 3D Game Creation

Details

• Google AI unveils Gemini 3, a next-generation large language model optimized for advanced code and graphics generation.
• A demo displays the model's ability to generate entire WebGL or Three.js projects, including textures, physics, and user interfaces, from a single natural-language prompt.
• The new structured code tokenizer increases the context window and reduces hallucinations in longer files, improving reliability in multi-file repositories.
• Internal benchmarks show a 17 percent accuracy boost over GPT-4 Turbo on HumanEval-X and a 22 percent gain on UnityBench coding tests.
• Gemini 3 is available now through the Gemini API on Google Cloud and within Bard, with a preview price of $0.002 per 1,000 input tokens and $0.006 per 1,000 output tokens.
• For enterprises, Gemini 3 features optional code-safety filters and on-premises key management adhering to Google’s Secure AI Framework (SAIF).

Impact

Gemini 3's launch amplifies competition with OpenAI and GitHub, pushing the boundaries of AI-assisted software development. Its ability to scaffold 3D games from a single prompt could democratize game prototyping and shorten development timelines. Google's lead in benchmarks and regulatory alignment highlights a shift towards safer, more specialized code-generation tools, raising the stakes for rival platforms.