Gemini 3 models: the short answer
For a general-purpose API evaluation, consider Gemini 3.8 Flash for complex agent workflows and long-horizon software engineering; Gemini 3.6 Flash for coding, knowledge work, and multimodal tasks; and Gemini 3.5 Flash-Lite or Gemini 3.1 Flash-Lite for high-volume, simpler requests. Gemini 3.1 Pro Preview is another candidate for demanding reasoning and precise, multi-step tool use, but it is a Preview model. These are workload-based starting points, not a universal ranking. Google’s API model pages and catalog distinguish Stable models from Preview; confirm the status and terms for the API channel you will actually use.
The shortlist below is for the Gemini Developer API. It does not imply equivalent availability through Vertex AI, Gemini apps, CLI tools, or other Google products.
General-purpose Gemini Developer API shortlist
| Model | API model ID | Catalog status | Selection starting point |
|---|---|---|---|
| Gemini 3.8 Flash | gemini-3.8-flash |
Stable | Google positions it for long-horizon software engineering, autonomous agents, and complex enterprise workflows. The model page lists a 1,048,576-token input limit and a 65,536-token output limit. Check the model page for the exact features your application needs. (Model page) |
| Gemini 3.6 Flash | gemini-3.6-flash |
Stable | Consider it for coding, knowledge work, and multimodal tasks. Google’s announcement reports improved output-token efficiency relative to Gemini 3.5 Flash; treat that as provider-reported evidence, not a guaranteed reduction on your workload. (Announcement) |
| Gemini 3.5 Flash-Lite | gemini-3.5-flash-lite |
Stable | Google Cloud’s Agent Platform documentation positions it for simple coding, document understanding, and lightweight agent workflows. That platform-specific guidance should not be assumed to describe identical behavior in every API channel. (Agent Platform documentation) |
| Gemini 3.1 Flash-Lite | gemini-3.1-flash-lite |
Stable | Google’s API documentation describes it for high-volume, latency-sensitive work such as simple extraction and lightweight agentic workflows. Check its model page for supported inputs and features. (Model page) |
| Gemini 3.1 Pro Preview | gemini-3.1-pro-preview |
Preview | Consider it for software engineering and agentic workflows requiring precise tool use and reliable multi-step execution. Google’s API page lists a 1,048,576-token input limit and a 65,536-token output limit; file search is listed as supported in AI Studio only. (Model page) |
The API catalog also lists specialized Live and text-to-speech models. They are not interchangeable with this general-purpose shortlist; consult their model-specific documentation for supported interactions and limits. (Gemini API model catalog)
Which Gemini 3 model fits your workload?
Complex agents or long-running coding tasks
Evaluate 3.8 Flash when your application needs multi-step planning, tool use, or sustained software-engineering work. Google describes it as its most intelligent Flash model, but that positioning does not prove it will outperform alternatives on your prompts. Test representative tasks.
Coding, knowledge work, or multimodal tasks
Include 3.6 Flash if those capabilities matter. Google reports improved token efficiency against 3.5 Flash, but the result is not a shared-condition comparison across every model or workload. Actual spend also depends on prompt and response size, tool calls, retries, and applicable rates.
High-volume, simpler requests
Compare 3.5 Flash-Lite and 3.1 Flash-Lite on representative classification, extraction, translation, or lightweight-agent tasks. The published positioning can help create a shortlist; it does not establish which is faster, cheaper, or more accurate for your application. In particular, do not transfer 3.5 Flash-Lite request guidance from Google Cloud’s Agent Platform to the Developer API without checking the relevant Developer API documentation.
Demanding reasoning and tool use
Evaluate 3.1 Pro Preview when complex reasoning, software engineering, or precise multi-step tool use is central. Its model page documents the separate gemini-3.1-pro-preview-customtools API endpoint for workflows combining bash and custom tools, and cautions that quality can fluctuate on tasks that do not benefit from those tools. Verify the model’s Preview terms before relying on it in production.
Gemini API pricing and cost trade-offs
The rates below are per million tokens. They are useful reference points, but they are not a complete workload estimate. The currently retrieved official pricing material did not establish matching, current standard paid rates for every model in this shortlist.
| Model | Published input / output rate | What the evidence establishes |
|---|---|---|
| 3.8 Flash | $0.75 / $3.75 introductory; $1.50 / $7.50 standard | Google’s pricing announcement says the introductory rate applies through December 31, 2026, with standard rates taking effect January 1, 2027. Recheck the live pricing page before budgeting. (Pricing announcement) |
| 3.6 Flash | $1.50 / $7.50 | Google’s announcement provides these rates as a reference, but the retrieved evidence does not confirm that they are the current standard rates. (Announcement) |
| 3.5 Flash-Lite | $0.30 / $2.50 | Google’s announcement provides these rates as a reference; current validity and any applicable pricing tiers were not established here. Do not use the figure as a current quote without checking official pricing. (Announcement) |
| 3.1 Flash-Lite | $0.25 / $1.50 | Google’s launch announcement gives these rates, but the retrieved evidence does not confirm current pricing. (Announcement) |
| 3.1 Pro Preview | Current rate not established | Do not substitute the older Gemini 3 Pro announcement rate: it covered a different model and specified prompts of 200,000 tokens or less. Confirm 3.1 Pro Preview’s current rate and any prompt-length tiers on the live pricing page. |
The retrieved material also did not establish whether thinking tokens are billed separately or how they are treated in each model’s output rate. Confirm that detail, prompt-length tiers, and any modality or service-tier charges before estimating cost. Google’s pricing page distinguishes paid access for production applications from free access for eligible models, and lists Batch API as a 50% cost-reduction option; check its current terms before applying those features to a budget. (Gemini Developer API pricing)
Production API checks before you commit
- Confirm model ID and status in the current Gemini API catalog; check Preview terms and deprecation notices.
- Verify features for the exact model and channel. Check supported inputs, tools, output formats, and limits in that model’s API documentation; do not assume family-wide parity.
- Build a cost estimate from current rates. Use your expected input and output volumes, and verify tiers, thinking-token billing, caching, batch use, and tool charges where applicable.
- Run a representative evaluation. Compare quality, completion rate, tool-call success, latency, and token use on your own tasks. Provider descriptions and benchmarks can inform what to test, but cannot predict your results.
- Check deployment access separately. Confirm status and terms for the Gemini Developer API or Vertex AI you intend to use. Access through an app, CLI, or IDE is not the same as API availability.
Quick shortlist
- Complex agent work or long-horizon coding: evaluate 3.8 Flash.
- Coding, knowledge work, or multimodal tasks: include 3.6 Flash.
- High-volume, simpler requests: compare 3.5 Flash-Lite and 3.1 Flash-Lite on your tasks and current rates.
- Demanding reasoning or precise multi-step tool use: evaluate 3.1 Pro Preview, accounting for its Preview status and verifying its current price.
Choose using measured results for your workload, not a model label or a single published rate.
Sources
What's new in Gemini 3.8 Flash
Introducing Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber
Gemini 3.6 Flash and Gemini 3.5 Flash-Lite documentation
Gemini 3.1 Flash-Lite announcement