Google Gemini Ultra 2.0 Beats GPT-5 on Multimodal Benchmarks, Can Process 10-Hour Videos
Google has unveiled Gemini Ultra 2.0 with native multimodal capabilities that outperform GPT-5 on video understanding, audio processing, and real-time reasoning tasks.
Google has officially released Gemini Ultra 2.0, its most powerful AI model yet, claiming significant performance advantages over OpenAI's GPT-5 on multimodal benchmarks.
Multimodal Superiority
Gemini Ultra 2.0's standout capability is its ability to process and reason over up to 10 hours of video content in a single query. This enables use cases ranging from meeting transcription to film analysis.
Benchmark Results
On the newly released VideoQA-Pro benchmark, Gemini Ultra 2.0 scores 94.2% versus GPT-5's 87.8%. On audio understanding tasks, the performance gap is even wider at 96.1% versus 84.3%.
Integration into Google Products
Gemini Ultra 2.0 powers the new Google AI Overviews feature in Search, which now provides 300-word summaries with cited sources for complex queries. It also powers the new NotebookLM Pro, which can process entire research libraries.
Pricing
Gemini Ultra 2.0 is available through Google AI Studio at $10 per million input tokens, significantly cheaper than comparable OpenAI models. Google One AI Premium subscribers at $19.99/month get access to the model through Gemini Advanced.
Related Articles
OpenAI Launches GPT-5 with Groundbreaking Reasoning Capabilities That Outperform Human Experts
OpenAI has unveiled GPT-5, its most powerful model yet, with advanced multi-step reasoning that outperforms human experts on 87% of professional benchmarks.
Anthropic Releases Claude 4 Opus with 2 Million Token Context Window — Sets New AI Benchmark
Anthropic's Claude 4 Opus sets a new benchmark with a 2 million token context window, enabling it to process entire codebases or research libraries in a single conversation.
DeepMind AlphaFold 3 Achieves 99% Accuracy on Protein Structures, Unlocking New Drug Targets
Google DeepMind's AlphaFold 3 has achieved 99% accuracy on known protein structures and can now predict how proteins interact with small molecules.
Google Pixel 9a Announced with Tensor G5 Chip and Industry-First 7-Year Update Guarantee
Google has unveiled the Pixel 9a featuring the Tensor G5 chip, 6.3-inch OLED display, and an industry-leading 7-year software update commitment at $499.