

Gemini 2.5
#8google · v2.5 · seit 2.5 Pro: ~März 2025 (Preview); GA 17. Juni 2025 | 2.5 Flash: 20. Mai 2025 (Preview); GA Juni 2025 | 2.5 Flash-Lite: 17. · 117× · zuletzt 30. Juni 2026
Gemini 2.5 ist eine multimodale KI-Modellfamilie von Google DeepMind, die als "Thinking Model" mit eingebauter Reasoning-Fähigkeit konzipiert ist. Sie umfasst u. a. die Varianten Pro, Flash und Flash-Lite und verarbeitet nativ Text, Bild, Audio, Video und Code über ein bis zu 1-Million-Token großes Kontextfenster. Die Modelle wurden im März 2025 als experimentelle Vorabversion vorgestellt und erreichten am 17. Juni 2025 die allgemeine Verfügbarkeit (GA); sie gelten laut Google als state-of-the-art bei Reasoning-, Coding- und Multimodalitäts-Benchmarks.</description> <parameter name="description_en">Gemini 2.5 is a multimodal AI model family from Google DeepMind, designed as a "thinking model" with built-in reasoning capability. It includes variants such as Pro, Flash and Flash-Lite, native
Features
| Key-Benchmark (%) | Humanity's Last Exam: 18,8% (state-of-the-art ohne Tool-Nutzung); GPQA Diamond: 84%; SWE-Bench Verified: 63,8% |
| Kontextfenster (Token) | 1.048.576 Token (1M), Output bis 64K Token |
| Lizenz | Proprietär (Closed-Source, kommerziell nutzungsbeschränkt) |
| Multimodalität | Nativ: Text, Bild, Audio, Video, Code; Output: Text (bzw. native Audio-Ausgabe je nach Variante) |
| Plattform | Google AI Studio, Gemini API, Vertex AI, Gemini App |
| Preis pro 1M Token | Gemini 2.5 Pro: $1,25 Input / $10,00 Output (Standard, <200K Kontext) |
| Release-Datum | Vorschau 25. März 2025 (I/O-Update 20. Mai 2025); GA 17. Juni 2025 |