Gemini 3 Pro vs GPT-5: the short answer
For a new production deployment, neither is a straightforward choice today: Google says the Gemini 3 Pro Preview API model (gemini-3-pro-preview) shut down on March 9, 2026. OpenAI lists GPT-5 (gpt-5) as deprecated and recommends moving to its latest model; its documentation does not give a GPT-5 shutdown date. This comparison is therefore most useful for reviewing legacy integrations, not as a recommendation to build on Gemini 3 Pro Preview.
Google’s Gemini 3 Pro Preview documentation lists text, image, video, audio, and PDF inputs. It also lists a larger input limit than the GPT-5 page: 1,048,576 tokens versus GPT-5’s maximum input of 272,000 tokens. That supports a comparison of the listed input limits, but the documentation used here does not establish whether Gemini supports more input modalities than GPT-5. Nor do these specifications establish a performance winner.
The exact API models and their status
The Google model here is Gemini 3 Pro Preview, API model code gemini-3-pro-preview—not Gemini 3.1 Pro Preview. Google’s model documentation says the Gemini 3 Pro Preview endpoint shut down on March 9, 2026.
The OpenAI model is GPT-5, API model ID gpt-5, with default snapshot gpt-5-2025-08-07. OpenAI currently labels it deprecated. Its documentation recommends the latest model but does not state a shutdown date. Deprecation should not be mistaken for confirmation that the endpoint is already shut down—or that it remains available to every account.
| API detail | Gemini 3 Pro Preview | GPT-5 |
|---|---|---|
| Model ID | gemini-3-pro-preview |
gpt-5 |
| Lifecycle status in provider documentation | Shut down March 9, 2026 | Deprecated; no shutdown date stated on the cited model page |
| Input and output limits listed | 1,048,576 input tokens; 65,536 output tokens | 272,000 maximum input tokens; 128,000 maximum output tokens; 400,000 context window |
| Input modalities listed in the documentation reviewed | Text, image, video, audio, and PDF | Not established in the model-specific passages used for this comparison |
| Function calling, structured outputs, and named tools | Not compared here | Not compared here |
These are provider-listed specifications, not guarantees of current availability for every API channel. Google marks Gemini 3 Pro Preview as shut down; OpenAI marks GPT-5 deprecated. For a workflow that depends on a particular modality, function-calling behavior, structured output, or tool, confirm support for the exact endpoint before relying on it. Limits and feature listings do not establish comparative accuracy, latency, or reliability.
Which model fit which production workload?
For long inputs, the documented limit favors Gemini 3 Pro Preview on paper. Its listed input limit was 1,048,576 tokens, compared with GPT-5’s listed maximum input of 272,000 tokens. The Gemini documentation also lists text, image, video, audio, and PDF inputs. That list describes Gemini’s documented inputs; the information reviewed here does not establish relative modality coverage versus GPT-5. Since the Gemini API model shut down on March 9, 2026, these differences matter for historical evaluations or migration planning—not as a live production choice.
For GPT-5, the documented limits are useful for reviewing an existing integration, but they do not establish an advantage in tool-driven workflows. The model is deprecated, and the specifications summarized here do not substantiate a comparison of function calling, structured outputs, or specific tools. If any of those are requirements, check their support for the exact API endpoint and plan around the provider’s recommended successor before relying on GPT-5 for a new deployment.
The comparison articles available for coding, RAG, and agent workflows discuss GPT-5.1 or GPT-5.2 variants, not base GPT-5. Their benchmark claims and recommendations should not be transferred to GPT-5. For a production decision, test the exact model IDs and representative tasks.
API pricing: historical comparison, not a current budget
The providers’ published rates allow a historical rate-arithmetic example for 200,000 input tokens and exactly 5,000 total billable output tokens:
| Calculation | Gemini 3 Pro Preview | GPT-5 |
|---|---|---|
| Input rate listed | $2.00 per 1 million tokens, for prompts of 200,000 tokens or less | $1.25 per 1 million tokens |
| Output rate listed | $12.00 per 1 million tokens | $10.00 per 1 million tokens |
| Input cost | 0.2 × $2.00 = $0.40 | 0.2 × $1.25 = $0.25 |
| Output cost, assuming 5,000 billed tokens | 0.005 × $12.00 = $0.06 | 0.005 × $10.00 = $0.05 |
| Estimated token total | $0.46 | $0.30 |
This calculation assumes the 5,000 output tokens are the total output quantity billed. It is not an estimate for a 5,000-token visible answer plus any additional output that may be billable. A visible answer of that length could therefore have a different total. The example excludes retries, tools, and other charges. Google announced the Gemini 3 Pro Preview rates as $2 per million input tokens and $12 per million output tokens for prompts of 200,000 tokens or less. OpenAI lists GPT-5 rates of $1.25 per million input tokens and $10 per million output tokens; it also lists cached GPT-5 input at $0.125 per million tokens, which is not included here.
Do not use these figures as a current production comparison. The Gemini endpoint used for the Google rate has shut down, and GPT-5 is deprecated. The arithmetic shows what the listed rates imply for the stated token counts; it does not establish that both APIs can currently serve the workload or what their replacements cost. API token charges are also separate from consumer app subscriptions.
How to make the production decision now
- Check the exact model ID in the deployed configuration. If it is
gemini-3-pro-preview, plan a migration: Google lists the endpoint as shut down. If it isgpt-5, treat it as deprecated and confirm current access rather than assuming a documented price makes it a suitable long-term dependency. - Choose a currently supported replacement, then evaluate that exact API model. Do not substitute successor-model specifications or prices into this GPT-5 versus Gemini 3 Pro comparison; replacements have different model IDs and need their own comparison.
- Test representative requests. Use the same prompt, input files, output requirements, and tool permissions where possible. Score task success and format validity, and record latency, tokens, retries, and total cost per completed task.
- Compare a router only if it solves a measured problem. A team can test routing by modality or task type, but should compare it with a single-model baseline using its own quality, latency, and cost measures. Routing is an option to evaluate, not a guaranteed improvement.
Bottom line: Gemini 3 Pro Preview’s documentation lists text, image, video, audio, and PDF inputs, and its documented input limit is larger than GPT-5’s listed maximum input limit. The evidence reviewed here does not establish that Gemini had broader modality coverage than GPT-5. GPT-5’s listed limits and historical prices can help with legacy-integration reviews, but the available evidence does not establish that it is better for tools or structured-output workflows. Gemini 3 Pro Preview’s API shut down on March 9, 2026, and GPT-5 is deprecated. Use this comparison to identify legacy dependencies and migration needs—not to choose between the two as though both were current, equivalent deployment options.
Sources
Gemini 3 Pro preview | Gemini API | Google AI for Developers
Gemini deprecations | Gemini API | Google AI for Developers
Gemini 3 for developers: New reasoning, agentic capabilities