Gemini 1.5
#52 v Frontier jazykové modelygoogle · v1.5 · od 15. Februar 2024 (Private Preview von Gemini 1.5 Pro); Mai 2024 Allgemeine Verfügbarkeit (GA) von Gemini 1.5 Pro und Fla · 26× · naposledy 30. 6. 2026
Gemini 1.5 is a family of frontier language models from Google, comprising the Pro, Flash, and Flash-8B variants, built on a sparse mixture-of-experts architecture. Its key feature is an extremely long context window of up to 1-2 million tokens, enabling native multimodality (text, image, audio, video) over very long inputs. At launch, Gemini 1.5 Pro achieved top scores on benchmarks such as MMLU, outperforming the predecessor Gemini 1.0 Ultra while using less compute. The models are available via Google AI Studio, the Gemini API, and Vertex AI, and are billed on a pay-per-token basis.
Vlastnosti
| Key Benchmark (%) | MMLU: 85.9% (5-shot), 91.7% (Majority Vote) – Gemini 1.5 Pro |
| Context Window (Tokens) | Up to 2,097,152 tokens (Pro); GA standard 1 million tokens |
| License | Proprietary (Google Gemini Terms of Use) |
| Multimodality | Text, image, audio, video (native multimodal input) |
| Platform | Google AI Studio, Gemini API, Vertex AI |
| Price per 1M Tokens | Pro: $1.25 input / $5.00 output (≤128k); Flash: $0.075 input / $0.30 output |
| Release Date | February 15, 2024 (Preview), GA in May 2024 |