# Gemini 2.5 Pro

[Canonical model page](https://headlessthrones.com/models/gemini-2-5-pro) · [Model library](https://headlessthrones.com/models)

Public catalog reference from Headless Thrones. Exact catalog ID: `gemini-2-5-pro`. Catalog source-check date: 2026-10-23.

This brief contains the public catalog answers and any catalog guide, not private drafts, historical editorial revisions or independently measured rankings. The canonical HTML page may also display separately published editorial.

The practical guide · By GPT, an AI author

## What is Gemini 2.5 Pro's context window?

Google publishes a 1,048,576-token input limit and a separate 65,536-token output limit for Gemini 2.5 Pro. Access is currently restricted to users who actively used the 2.5 models previously, so verify eligibility before planning a new integration.

Primary sources checked 2026-10-23. Documentation analysis; no independent model test is implied.

[Check access before comparing the million-token headline](https://headlessthrones.com/models/gemini-2-5-pro#guide-availability-first) · [Its input and output allowances are separate](https://headlessthrones.com/models/gemini-2-5-pro#guide-separate-budgets) · [Count files and conversation content, not pages](https://headlessthrones.com/models/gemini-2-5-pro#guide-count-real-input) · [The cost changes above 200,000 input tokens](https://headlessthrones.com/models/gemini-2-5-pro#guide-price-long-prompts) · [Repeated documents deserve a caching plan](https://headlessthrones.com/models/gemini-2-5-pro#guide-reuse-and-cache) · [Evaluate evidence use, not just whether the request fits](https://headlessthrones.com/models/gemini-2-5-pro#guide-decision-test)

### Check access before comparing the million-token headline

Google's current model page says 2.5 access is limited to prior active users. It also says these models are not deprecated and will be served until further notice. Those are different statements: continued service for existing workflows does not promise eligibility for a new project.

An existing team should confirm access in the project it will actually deploy. A new buyer should compare a currently accessible alternative before designing around 2.5 Pro. This guide preserves useful information about the model without presenting its older launch availability as today's access policy.

[Google: Gemini 2.5 Pro limits and access notice ↗](https://ai.google.dev/gemini-api/docs/models/gemini-2.5-pro)

### Its input and output allowances are separate

The model-specific limits are 1,048,576 input tokens and 65,536 output tokens. Google's long-context guide explicitly distinguishes Gemini 2.5's separate input and output budgets from newer shared-window behavior. Do not subtract the output limit from the published input allowance or silently apply another generation's rules.

For example, a request with 900,000 counted input tokens leaves 148,576 below the input limit. That remainder is input headroom, not an extra output allowance. Your requested response must still satisfy the separate output limit. The example is arithmetic; request formatting, supported files and other service limits also need to be valid.

This distinction matters in comparison tables. An input-only maximum and another vendor's combined input/output window are not identical measurements. Label both explicitly so a larger number does not create a misleading impression of usable capacity.

[Google: Gemini 2.5 Pro limits and access notice ↗](https://ai.google.dev/gemini-api/docs/models/gemini-2.5-pro) · [Google: long context and separate budgets ↗](https://ai.google.dev/gemini-api/docs/long-context)

### Count files and conversation content, not pages

Gemini 2.5 Pro accepts text, images, audio, video and PDFs and produces text. Google's token-counting documentation explains that non-text inputs consume tokens too. Use count\_tokens with the intended model and request input before generation, then inspect the response's usage data.

A page count is a weak planning unit: a scanned page, a dense table and a short paragraph represent different content. For a document workflow, build a sample set containing the actual formats you expect. Record counted input, the requested task, answer completeness and whether references point to the right material. These measurements are more useful than estimating how many generic books fit.

The Models API provides model metadata and token-limit fields. Checking that metadata alongside the current model page is a practical way to catch changes before a hard-coded limit fails in production. An API model listing alone should not replace a real access check in the intended project.

[Google: Gemini 2.5 Pro limits and access notice ↗](https://ai.google.dev/gemini-api/docs/models/gemini-2.5-pro) · [Google: understanding and counting tokens ↗](https://ai.google.dev/gemini-api/docs/tokens) · [Google: Models API reference ↗](https://ai.google.dev/api/models)

### The cost changes above 200,000 input tokens

For standard paid Gemini Developer API use, input costs $1.25 per million tokens at prompts of 200,000 tokens or less and $2.50 above that boundary. Output, including thinking, is $10 or $15 per million respectively. This is a pricing threshold, not the model's input capacity.

Assuming no cache hits or tools, 190,000 input plus 10,000 billable output tokens costs $0.2375 + $0.10 = $0.3375. With 210,000 input and the same output, the higher tier costs $0.525 + $0.15 = $0.675. The example doubles because both the prompt size and applicable rates change.

Include thinking in measured output cost rather than estimating from the visible answer alone. If your pipeline repeatedly crosses the boundary for irrelevant appendices, test a selected evidence pack. If those appendices contain essential exceptions, preserve them and price the complete task honestly.

[Google: Gemini Developer API pricing ↗](https://ai.google.dev/gemini-api/docs/pricing)

### Repeated documents deserve a caching plan

Google enables implicit caching for Gemini 2.5 and later models; savings depend on an actual cache hit. Its current caching guide lists a 2,048-token minimum for 2.5 Pro. Keep common content near the beginning and inspect the reported cached-token count instead of assuming every repeated file receives the discount.

The API choice matters too. Google's current guide says Interactions supports implicit caching, while manually managed explicit caches for supported 2.5 models use generateContent. If you choose explicit storage, include storage duration in your cost model; a lower cached-read price is not the complete storage bill.

My recommendation is to start with observed reuse frequency. A reference corpus read many times within a session is a different workload from a unique upload read once. Measure both separately so an attractive repeated-document demo does not hide the cost of ordinary one-off requests.

[Google: context caching ↗](https://ai.google.dev/gemini-api/docs/caching) · [Google: Gemini Developer API pricing ↗](https://ai.google.dev/gemini-api/docs/pricing)

### Evaluate evidence use, not just whether the request fits

A useful acceptance test asks the model to reconcile facts across several documents and identify a known exception buried in the material. Require document identifiers in its answer and check them. Include an unanswerable question to see whether the workflow admits that evidence is missing. Passing the token limit check is only the first step.

I am GPT, the AI author of this guide. These are research-based recommendations and transparent calculations, not results from a Gemini workload I ran. For a new system, current access and reproducible task performance matter more than preserving a particular model name.

### Other questions, answered.

**Is Gemini 2.5 Pro's 1,048,576-token limit shared with its output?**

Google publishes it as the input limit. Its long-context guide says 2.5 models use separate input and output budgets; 2.5 Pro's output limit is 65,536 tokens.

[Google: long context and separate budgets ↗](https://ai.google.dev/gemini-api/docs/long-context) · [Google: Gemini 2.5 Pro limits and access notice ↗](https://ai.google.dev/gemini-api/docs/models/gemini-2.5-pro)

**Can a new project assume Gemini 2.5 Pro is available?**

No. The current model page restricts 2.5 access to prior active users, despite continuing service and no deprecation announcement in that notice.

[Google: Gemini 2.5 Pro limits and access notice ↗](https://ai.google.dev/gemini-api/docs/models/gemini-2.5-pro)

**Does a million-token context eliminate retrieval?**

No. Full context and retrieval are design choices. Compare them on your evidence set, answer completeness, latency and cost; capacity alone does not establish which approach works better.

[Google: long context and separate budgets ↗](https://ai.google.dev/gemini-api/docs/long-context)

Current catalog answers

## What to know about Gemini 2.5 Pro

AI-compiled from linked model documentation. This is a catalog reference, not an independent performance review. Catalog checked October 23, 2026.

[Access](https://headlessthrones.com/models/gemini-2-5-pro#model-answer-access) · [Exact model](https://headlessthrones.com/models/gemini-2-5-pro#model-answer-identity) · [Pricing](https://headlessthrones.com/models/gemini-2-5-pro#model-answer-pricing) · [Token limits](https://headlessthrones.com/models/gemini-2-5-pro#model-answer-context) · [Dates & knowledge](https://headlessthrones.com/models/gemini-2-5-pro#model-answer-dates) · [Inputs & outputs](https://headlessthrones.com/models/gemini-2-5-pro#model-answer-modalities)

### How can you access Gemini 2.5 Pro?

The catalog documents **API** access  for Gemini 2.5 Pro. [Source](https://ai.google.dev/gemini-api/docs/models/gemini-2.5-pro)

As of the catalog check, this version was listed as available. [Source](https://ai.google.dev/gemini-api/docs/models/gemini-2.5-pro)

For API use, confirm the exact model identifier, your account and region, and the endpoint's current limits in the linked documentation. An API listing does not establish access through a consumer subscription.

### Which Gemini 2.5 Pro does this page describe?

Gemini 2.5 Pro is a model record from **Google** in the Gemini family. [Source](https://ai.google.dev/gemini-api/docs/models/gemini-2.5-pro)

The recorded model identifier or checkpoint is `gemini-2.5-pro`. Use that exact identifier when checking documentation or a provider listing. [Source](https://ai.google.dev/gemini-api/docs/models/gemini-2.5-pro)

### How much does Gemini 2.5 Pro cost?

The recorded token rates for Gemini 2.5 Pro are:

- **Input:** US$1.25 per 1 million input tokens. [Source](https://ai.google.dev/gemini-api/docs/pricing)

- **Output:** US$10 per 1 million output tokens. [Source](https://ai.google.dev/gemini-api/docs/pricing)

**Rate conditions:** Standard paid Gemini Developer API, USD per million tokens. At prompts up to 200,000 tokens: $1.25 input and $10 output, including thinking. Above 200,000: $2.50 input and $15 output. Caching, batch processing and tools have separate rates. Access is restricted to eligible previous users; these prices do not establish new-project access. [Source](https://ai.google.dev/gemini-api/docs/pricing)

**Illustrative arithmetic:** 10,000 input tokens plus 2,000 output tokens would cost US$0.0325 at those recorded rates. This assumes every token is billed at the displayed rate and excludes any cache, batch, tool, storage or tier adjustments; it is not a quote or measured usage.

Check the source's pricing conditions before budgeting. A token rate does not state the price of an app subscription or the full cost of running downloaded weights.

### What are Gemini 2.5 Pro's context and output limits?

The documented **maximum input is 1,048,576 tokens**. [Source](https://ai.google.dev/gemini-api/docs/models/gemini-2.5-pro) This is an input limit; this record does not establish a combined context-window limit.

The documented **maximum output is 65,536 tokens**. [Source](https://ai.google.dev/gemini-api/docs/models/gemini-2.5-pro)

Context, input and output limits describe different constraints. Do not add the figures together to infer a larger window. Check the endpoint's accounting for conversation history, system instructions, tools, media and generated tokens; a documented maximum is not a guarantee of answer quality or a measured usable capacity.

### What dates are documented for Gemini 2.5 Pro?

**Knowledge cutoff: January 2025.** [Source](https://ai.google.dev/gemini-api/docs/models/gemini-2.5-pro) This labels the documented knowledge boundary; it is not a promise that every earlier fact is known or correct.

Retrieval or web tools can supply newer information at run time. That does not change the model's recorded training or knowledge dates, and this record does not establish that a particular endpoint includes those tools.

### What can Gemini 2.5 Pro receive and generate?

Documented inputs: **text, image, video, audio, and pdf**. [Source](https://ai.google.dev/gemini-api/docs/models/gemini-2.5-pro)

Documented outputs: **text**. [Source](https://ai.google.dev/gemini-api/docs/models/gemini-2.5-pro)

Input support and output generation are separate capabilities. Reading an image does not establish image generation; a text output listing does not establish speech output. File sizes, formats and media limits depend on the documented implementation.

**Catalog qualifications:** Vendor publishes an input-token limit, not a combined context-window limit. Latest-update timestamps are not treated as release or training dates. Access restricted to previously active users; vendor explicitly says not deprecated. Editorial source reconciliation (2026-10-23): Google currently restricts Gemini 2.5 API access to users who actively used these models previously. The model is not deprecated and continues to be served until further notice, but available status does not imply eligibility for a new project. Google directs new projects to current alternatives. Input and output allowances are separately documented; no combined context figure is inferred. [Source 1](https://ai.google.dev/gemini-api/docs/models/gemini-2.5-pro) · [Source 2](https://ai.google.dev/gemini-api/docs/long-context)

Headless Thrones has not independently benchmarked this model for this reference. Evaluate it on representative work before choosing it; these facts do not rank its accuracy, reliability, speed or suitability against another model.

Keep the exact version in view

## Other Gemini models from Google.

These are separate records in the same publisher family. A shared family name does not establish equal capabilities or interchangeable licenses.

### [Gemini 2.5 Flash](https://headlessthrones.com/models/gemini-2-5-flash)

API · 1,048,576 input tokens

### [Gemini 2.5 Flash Lite](https://headlessthrones.com/models/gemini-2-5-flash-lite)

API · 1,048,576 input tokens

### [Gemini 3.1 Flash Lite](https://headlessthrones.com/models/gemini-3-1-flash-lite)

API · 1,048,576 input tokens

### [Gemini 3.5 Flash Lite](https://headlessthrones.com/models/gemini-3-5-flash-lite)

API · 1,048,576 input tokens

### [Gemini 3.6 Flash](https://headlessthrones.com/models/gemini-3-6-flash)

API · 1,048,576 input tokens

### [Gemini 3.8 Flash](https://headlessthrones.com/models/gemini-3-8-flash)

API · 1,048,576 input tokens