Skip to content
News archive
ArchiveJune 18, 2025

Google Gemini 2 Ultra: New Features That Challenge GPT-4

Google's latest Gemini 2 Ultra model introduces groundbreaking capabilities in code generation, scientific reasoning, and multimodal tasks that directly compete with OpenAI's offerings.

Google has released Gemini 2 Ultra, its most powerful AI model to date, introducing several features that position it as a serious competitor to GPT-4 and Claude 3.5 Sonnet.

What's New in Gemini 2 Ultra

Advanced Code Generation: Gemini 2 Ultra achieves state-of-the-art performance on coding benchmarks, including HumanEval and SWE-bench. It can now generate, debug, and refactor code across 50+ programming languages with near-perfect accuracy on standard tasks.

Scientific Reasoning: The model demonstrates remarkable capabilities in scientific problem-solving, including molecular modeling, physics simulations, and mathematical proofs. Google reports it outperforms GPT-4 on graduate-level science exams by 15%.

Native Tool Use: Unlike previous versions that required external integrations, Gemini 2 Ultra can natively browse the web, execute code, and interact with Google Workspace tools within a single conversation flow.

2M Token Context: The model supports a 2 million token context window, the largest of any commercial LLM, enabling it to process entire repositories or multi-volume document sets.

Pricing and Access

Gemini 2 Ultra is available through Google AI Studio and the Vertex AI platform. Pricing starts at $0.005 per 1K input tokens, making it competitive with GPT-4 Turbo while offering a larger context window.

Developer Reactions

Early adopters report impressive results, particularly in code generation and data analysis tasks. However, some note that GPT-4 still maintains an edge in creative writing and nuanced conversation.