Gemini 2.5 Pro — Specifications
| Desenvolvedor | Google (Google DeepMind) |
|---|---|
| Tipo | LLM (frontier reasoning / thinking) |
| Modalidade | Text, Image, Audio, Video, PDF → Text |
| Parâmetros | Não divulgado |
| Janela de contexto | 1M (1,048,576 tokens) |
| Saída máxima | 64K (65,536 tokens) |
| Licença | Proprietário |
| Pesos abertos | ❌ Não |
| Lançado | March 2025 (preview); June 17, 2025 (GA) |
| Preço por entrada | $1.25 /1M |
| Preço por saída | $10 /1M |
| Provedores de API | Google AI Studio (Gemini API), Vertex AI, OpenRouter |
Gemini 2.5 Pro is Google DeepMind’s flagship “thinking” model from the 2.5 generation, built to reason through problems before it answers. It debuted as an experimental preview on March 25, 2025 and reached general availability on June 17, 2025, quickly establishing itself as a benchmark leader in coding, mathematics, and long-context reasoning.
Its headline feature is a native 1,048,576-token (~1M) context window paired with fully multimodal input — text, images, audio, video, and PDFs — and text output up to 65,536 tokens. Adaptive “thinking budgets” let developers trade latency and cost for deeper reasoning. The model is served through Google AI Studio, the Gemini API, and Vertex AI, with third-party access via OpenRouter.
Pricing is tiered by prompt size: $1.25 per million input tokens up to 200K, rising to $2.50 above that, with output at $10 and $15 respectively.
Veredito: Gemini 2.5 Pro remains one of the strongest value picks among frontier reasoning models — a genuine 1M-token multimodal context at mid-tier pricing. It is closed-weight and API-only, but for long-document analysis, agentic coding, and mixed-media workloads it is hard to beat. Newer Gemini 3.x models now supersede it, yet 2.5 Pro stays widely deployed and well-supported.
