Gemini 2.5 Pro: Google's Thinking Model

Google has released Gemini 2.5 Pro, their most capable model to date, featuring built-in thinking capabilities that activate automatically when the model encounters complex problems. The model supports a 1-million-token context window and achieves strong scores across reasoning, coding, and knowledge benchmarks.

On MMLU, Gemini 2.5 Pro scores 92.0, placing it among the top models globally. Its GPQA Diamond score of 68.4 demonstrates solid graduate-level reasoning, though it trails some specialized reasoning models. Where the model shines is in its combination of long context, multimodal understanding, and thinking capability in a single package.

The thinking feature works by allowing the model to generate internal reasoning traces before producing its final response. Unlike some reasoning models that expose these traces to users, Gemini 2.5 Pro's thinking is largely internal, with the API providing a "thinking" field that developers can optionally access.

Pricing is competitive at $1.25 per million input tokens (under 200K tokens) and $10 per million output tokens. The model is available through Google's Vertex AI platform and through third-party providers like OpenRouter.

For developers already in Google's ecosystem, Gemini 2.5 Pro offers a compelling combination of capability and convenience. The 1M context window is particularly valuable for applications that need to process large documents or maintain extensive conversation histories.