The latest evaluations of AI coding models reveal that Gemini 2.5 Pro continues to lead in coding performance and cost efficiency compared to OpenAI's o3 and o4-mini models. While o3 scored 6.7% higher than Gemini 2.5 Pro in some benchmarks, it is approximately 17 times more expensive, and o4-mini scored lower than Gemini 2.5 Pro despite being three times more costly. Gemini 2.5 Pro also maintains a higher pass rate on coding tasks on the first try, only being surpassed by o3 after multiple attempts. Additionally, Gemini 2.5 Pro outperforms o4-mini on long context benchmarks, handling up to 1 million tokens compared to o4-mini's 130,000 tokens. In terms of coding style, Gemini 2.5 Pro respects original code structure more than o4-mini, which tends to rewrite existing code. OpenAI's o3 and o4-mini excel in refining specifications and building new features, with o3 offering strong improvement suggestions. On broader benchmarks such as the Extended NYT Connections and Thematic Generalization, o3 and o4-mini rank well but remain slightly behind other models like o1-pro and Claude 3.7 Sonnet Thinking 16K. Overall, Gemini 2.5 Pro holds a balance of strong coding ability, cost-effectiveness, and long-context handling that currently positions it ahead of OpenAI's comparable models.