Gemini 2.5 Pro
Gemini 2.5 Pro by Google DeepMind is a thinking model with sparse MoE architecture, 1M-token context, and native multimodal input across text, audio.
Model Overview
Capabilities, design details, and architectural traits
Gemini 2.5 Pro - Google's Thinking Model with Sparse MoE
Gemini 2.5 Pro is a thinking model from Google DeepMind, meaning it reasons through its thoughts before responding. This behavior is built into the model's training, not applied as a post-processing step. Its architecture is a sparse mixture-of-experts (MoE) transformer, which activates only a subset of parameters per input token - decoupling total model capacity from per-token compute cost.
Documented Traits That Define Gemini 2.5 Pro
| Trait | Detail |
|---|---|
| Thinking model | Reasons before responding; thinking capability is built into training via reinforcement learning and improved post-training |
| Sparse MoE architecture | Dynamically routes each input token to a learned subset of parameters (experts); total capacity is decoupled from serving cost per token |
| 1M-token context window | Accepts text, audio, images, video, and entire code repositories within a single 1M-token context; outputs up to 64K tokens |
| Thought summaries | Raw model thoughts are structured into a formatted output with headers and key details, available via the Gemini API and Vertex AI |
| Deep Think mode | An enhanced reasoning variant of 2.5 Pro that uses parallel thinking - the model considers multiple hypotheses before responding |
Native Multimodality as a Core Input Design
Gemini 2.5 Pro accepts text, audio, images, video, and code repositories natively within its context window. The model card explicitly lists entire code repositories as a supported input type, positioning the 1M-token window as a practical tool for agentic coding workflows rather than only document processing.
Benchmark Performance
Independent evaluations · Artificial Analysis
Accuracy & Capability Details
Compare Models Side-by-Side
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | Anthropic | Anthropic | |
| Release Date | June 5, 2025 | September 22, 2026 | September 28, 2026 |
| Knowledge Cutoff | Jan 2025 | - | Jun 2026 |
| Context & Limits | |||
| Context Window | 1.0M Best Context Window | 1M | 1M |
| Pricing (per 1M tokens) | |||
| Input Pricing | $1.25 Best Input Pricing | $4 | $2 |
| Output Pricing | $10 Best Output Pricing | $20 | $10 Best Output Pricing |
| Modalities | |||
| Inputs | textimagefileaudiovideo | textimagefile | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 16.1 | 57.6 Best Intelligence Index | 56.0 |
| Coding Index | 33.3 | - | - |
| Agentic Index | 1.6 | - | - |
Humanity's Last Exam
Extremely difficult logical reasoning and knowledge.
Long Context Reasoning
Logical reasoning over long context windows.
SciCode Benchmark
Scientific coding and mathematical modeling.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
Explore more from Google
Other models by Google
Gemini 4 Argon
Gemini 3.8 Flash
Gemini 3.7 Flash
Gemini 3.6 Flash
Top AI Models
Leading alternatives by intelligence score
