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 is Google's family of hybrid reasoning models (Pro, Flash, Flash-Lite) with native multimodality supporting text, image, audio, and video input. The models feature controllable "thinking" capabilities via configurable thinking budgets and a 1-million-token context window. Gemini 2.5 Pro is the most capable model for complex reasoning and coding; 2.5 Flash is optimized for price-performance balance; 2.5 Flash-Lite is the most cost-efficient model in the family for high-throughput workloads. All three models are available via the Gemini API (Google AI Studio) and Vertex AI (Google Cloud).
Fonctionnalités
| Key Benchmark (%) | Humanity's Last Exam: 18.8% (state-of-the-art without tool use); GPQA Diamond: 84%; SWE-Bench Verified: 63.8% |
| Context Window (Tokens) | 1,048,576 tokens (1M), output up to 64K tokens |
| License | Proprietary (closed-source, commercial usage restricted) |
| Multimodality | Native: text, image, audio, video, code; output: text (or native audio output depending on variant) |
| Platform | Google AI Studio, Gemini API, Vertex AI, Gemini App |
| Price per 1M Tokens | Gemini 2.5 Pro: $1.25 input / $10.00 output (standard, <200K context) |
| Release Date | Preview March 25, 2025 (I/O update May 20, 2025); GA June 17, 2025 |
