Gemini 3 Pro Versions Compared: Context, Modality, Price

Verification date: 2026-09-21. The catalog fields and prices in this article are converted from the public response of https://bazaarlink.ai/api/v1/models on that date. Prices change, so follow the official site and the latest BazaarLink announcements.

When searching for "gemini 3 pro" or "gemini3 pro", the easiest mistake is to treat the product name, the model ID in the official documentation, and the IDs still selectable in a third-party model catalog as the same thing. This article first separates these names, then uses the versions actually listed in the catalog on that day to compare context length, modality, reasoning and token prices.

Summary: First, Clarify What "Gemini 3 Pro" Refers To

As of 2026-09-21, the Pro mainline listed in the official Gemini 3 developer documentation is gemini-3.1-pro-preview. The official changelog also states that the old Gemini 3 Pro Preview was shut down on 2026-03-09 and points to gemini-3.1-pro-preview. Therefore, articles, code and configuration files should not just say "Gemini 3 Pro" without confirming the actual ID.

When I queried the BazaarLink public model API this time, it did not return a text model with the exact name of the Pro version, but it did return gemini-3.1-pro-preview, gemini-3.1-pro-preview-customtools and gemini-3-pro-image. These three names look similar, but their output modalities and uses differ.

Which Gemini Versions Were Actually in the Catalog on 2026-09-21

The table below directly organizes the public API response. The raw prompt and completion values are USD real numbers per token; here they are multiplied by 1,000,000 to convert them into an easier-to-read "USD per 1M tokens". reasoning is a summary of the field returned by the API; "not provided" means the field is null, not an inference that it is supported or unsupported.

This table is a "catalog snapshot for that day", not a permanent guarantee. Previews, aliases, deprecations and new versions can all change, so re-read the model list before any actual call.

Pro, Customtools and Image: How to Read the Differences

gemini-3.1-pro-preview is the text/multimodal Pro version closest to what people usually mean when they search for "Gemini 3 Pro": 1M context, inputs that can include text, images, files, audio and video, and text output. The reasoning field in the catalog is marked mandatory=true, which means it should not be compared as an ordinary Flash model that "definitely does no reasoning".

The context, modality, reasoning and token prices of gemini-3.1-pro-preview-customtools are the same in this catalog response. Its difference lies in the use case of the custom tools endpoint implied by its ID. If your application needs tool calls, base the decision on the actual API interface and your own documentation-based test results. Do not assume the two behave identically in every respect just because their prices are the same.

gemini-3-pro-image is the image-output route: the catalog lists it as text+image→text+image, its context is only 131,072, and its completion price is likewise $12 per 1M. However, its output tokens and image output should not be interpreted with the same logic. The official pricing page also lists it separately from text/thinking output, with image output carrying a separate per-image equivalent price.

Choosing by Task: Long Context, Vision, Reasoning or Cost

If the work is long-document analysis or codebase understanding, first check whether the 1M context and the input modalities truly match your data format, and then look at output cost. Context length is the upper limit that can be held, not a sign that every request will use it all, and it is not equivalent to a monthly quota.

If the need is to generate or edit images, first check whether the output is text+image. Do not put gemini-3-pro-image and the text-output Pro version into the same "which one is smarter" question. Image work also requires confirming sizing, per-image billing and API endpoint rules. This article does not turn rules that do not appear in the catalog fields into conclusions.

If the need is controllable reasoning depth, then look at default_effort, supported_efforts or mandatory within reasoning. The catalog this time does not provide the quality, latency or measured accuracy of each model, so these items are listed as unverified. A ranking cannot be derived from price or model year.

To view the fields for that day directly, start with the BazaarLink AI model catalog, then use the API documentation to compare integration methods. For SDK examples, use the SDK documentation.

How to Read Prices, and Why Not to Look Only at the Model Name

Using the catalog figures for gemini-3.1-pro-preview as an example: 1M prompt tokens cost $2, and 1M completion tokens cost $12. If a month has 2M prompt tokens and 0.5M completion tokens, a pure per-token calculation gives 2 × 2 + 0.5 × 12 = $10. Actual billing must still follow the public prices of that day, the request type and the latest announcements. This example does not add caching, tools, per-image billing or other possible billing fields on its own.

With the same "Pro" label, gemini-2.5-pro, gemini-2.5-pro-preview and gemini-2.5-pro-preview-05-06 have identical input and output prices in this catalog, but their modalities and IDs are not entirely the same. When selecting a model, record the actual ID, input format, output format and reasoning field together, so that price or version changes can be traced later.

Pre-Use Checklist and Next Steps

1. Send a new GET to the public model list to confirm that the ID you plan to use is still listed.
2. Confirm whether your input data is text, images, files, audio or video, and check it against architecture.modality.
3. Confirm the context limit, reasoning settings and output type, and do not just copy the product name from search results.
4. Estimate monthly cost separately for prompt and completion, and record the date of the estimate.
5. Before going live, read the official model and pricing pages again. All prices in this article are the public prices of 2026-09-21 converted; when the official site or this site updates, follow the latest announcement.

If you already know which ID you want to connect, you can start from the model catalog to confirm the fields, then build your request from the developer documentation. The conclusions of this article only go as far as model selection and cost estimation, and do not present unverified quality or performance as guarantees.

Verification sources

• Public model API: https://bazaarlink.ai/api/v1/models
• Model catalog: https://bazaarlink.ai/models
• Official Gemini 3 developer guide: https://ai.google.dev/gemini-api/docs/gemini-3
• Official Gemini API pricing: https://ai.google.dev/gemini-api/docs/pricing
• Official Gemini API changelog: https://ai.google.dev/gemini-api/docs/changelog

TWD billing · Taiwan invoices · leading AI models · OpenAI-compatible API