Gemini 2.5 Pro
Google's Native Reasoning Frontier
Google DeepMind's first model to integrate reasoning natively into its main product line — earlier reasoning attempts (Gemini 2.0 Flash Thinking, December 2024) were experimental side products. Gemini 2.5 Pro became the default Gemini for paying users in March 2025 and held a meaningful lead on math reasoning benchmarks (AIME, Humanity's Last Exam) for ~2 months.
Intelligence
Below avg
Speed
Medium
115 tok/s output
Cost
High
$1.25 in / $10.00 out
Context
1M
Up to 1,000,000 tokens
How are Intelligence, Speed & Cost bucketed?
Intelligence and Speed buckets are percentile ranks on
Artificial Analysis. Cost buckets are fixed dollar
thresholds keyed off output-token price ($/M out).
Intelligence
- Top 1%≤ 1%
- Top 5%≤ 5%
- Top 10%≤ 10%
- Good≤ 25%
- Medium≤ 50%
- Below avg> 50%
Speed
- Top 1%≥ 345 tok/s
- Top 5%≥ 237 tok/s
- Top 10%≥ 196 tok/s
- Good≥ 146 tok/s
- Medium≥ 90 tok/s
- Slow< 90 tok/s
Cost
- Freeopen weights · self-host
- Low< $1 / M out
- Moderate$1–5 / M out
- High≥ $5 / M out
Why it matters
Gemini 2.5 Pro proved Google could ship at the frontier on Google's calendar — not just react to OpenAI. The combination with TPU v6 / v7 manufacturing capacity gave Google unique cost-structure advantages that became visible in 2025-26 pricing.
Core Capabilities
Long Documents
Handles entire codebases, books, and multi-doc RAG.
Multimodal
Combines text, vision, and audio in one model.
Reasoning
Solves complex math, logic, and planning tasks.
Agent Workflows
Built for tool use and autonomous tasks.
Context Window
1M tokens
≈ entire codebase
4k Chat 聊天
32k Long docs 长文档
128k Books 整本书
400k Multi-doc 多文档
1M This model 本模型
10M
Availability
API
Available
Product / App
Available
Open Source
Not released
Enterprise
Available
Pricing Model
Pay per token
Input and output billed separately.
Pay-per-token Capability / Performance
Where this model sits relative to the middle 60% of models in the tree. All scores are 0–10 (higher is better).
Lower 20% Upper 80% This model
Context / memory
Context window size · log-scaled
6.0
9.0
9.0
Lower 20% 20th percentile — 20% of models score below this This model Where the current model lands Upper 80% 80th percentile — only 20% of models score above this
Percentile boundaries are computed across every model in the tree that reports the underlying benchmark for each capability.
What it feels like
- Long context is the headline — 91.5% on 128k and 83.1% on 1M long-context retrieval, ahead of every other frontier model
- Independent testers confirm reliable retrieval up to ~800K tokens; the last 200K degrades
- Genuine 'put a whole codebase or book in the prompt' workflow — not theatre
- AIME 2024 92.0% and GPQA Diamond 84.0% put it in the top tier on reasoning
- Built-in thinking mode is a default, not a switch — you don't have to pick a 'reasoning model'
- Pricing tiers up past 200K tokens (2x surcharge) — long-context isn't free
Best use cases
- Whole-codebase analysis, refactors across many files at once
- Long-form document QA (books, contracts, research papers)
- Multi-doc RAG replacement when context budget is generous
- Mixed reasoning + multimodal tasks (still strong on MMMU, MMLU-Pro)
Tools to try
Not ideal for
- Pure agentic-coding leaderboards — Claude Opus 4.5 and DeepSeek V4 lead SWE-bench
- Cost-bound bulk inference past 200K tokens (price doubles)
- Self-hosted / open-weights deployments
Model Evolution
Gemini 1
Dec 2023
Gemini Robotics
Mar 2025
Gemma 3
Mar 2025
Gemini 2.5 Pro
Mar 2025
Gemini 3 Pro
Nov 2025