Google’s Gemini 2.5 Pro Outperforms OpenAI’s O3 Model in Handling Complex, Lengthy Texts

Google’s Gemini 2.5 Pro currently outperforms other models in processing complex, lengthy texts, as evidenced by the Fiction.Live benchmark. This test evaluates a model’s ability to understand and convey intricate narratives, surpassing simpler tasks like search functions. OpenAI’s o3 model matches Gemini’s performance for contexts up to 128,000 tokens but declines significantly at 192,000 tokens, whereas Gemini maintains over 90% accuracy even at that length. While Gemini claims a maximum of one million tokens, its accuracy may decrease with longer contexts. In contrast, OpenAI’s o3 has a 200,000-token limit. Meta’s Llama 4 Maverick offers up to ten million tokens but struggles with long-context intricacies. Google DeepMind’s Nikolay Savinov highlights that larger contexts can lead to diminished attention on each token, recommending a selective approach to information. Consequently, users should eliminate irrelevant content when utilizing models for lengthy documents to improve performance and reasoning abilities.

Source link