Start Google API volume evaluations with Gemini 3.8 Flash. Its introductory rates run through December 31, 2026, with higher published rates from January 1, 2027. Recheck compatibility and quality when migrating.
Gemini 4 Argon is now available on the paid Gemini API as gemini-4-argon. Its introductory Standard rates double December 1, after the November 30 cutoff. Caching is implicit only; the API Free tier does not include Argon. Check its model detail page for current context and output limits; long decode and synchronous output are different execution modes. Read the refreshed Argon pricing and access analysis.
The chart, calculator, and table above read pricing.json at build time. Use the Google AI pricing page for the entire tracked catalog or the full AI pricing table for cross-provider comparisons.
Which Gemini model should you use?
| Workload | Start with | Escalate when | Measure |
|---|---|---|---|
| Classification, extraction, routing | Gemini 3.1 Flash-Lite | Validation or nuance fails | Valid records per dollar |
| Support, product Q&A, routine RAG | Gemini 3.8 Flash | Answer quality misses the target | Resolved requests per dollar |
| Multimodal app flows | Gemini 3.8 Flash | Complex interpretation needs Pro | Accepted outputs and latency |
| Premium RAG and document analysis | Gemini 2.5 Pro | Newer preview behavior is required | Supported claims per dollar |
| Frontier reasoning evaluation | Gemini 4 Argon versus Pro Preview | Accepted-task gains justify spend | Correct decisions per dollar |
| Long-context review | A Pro route after tier modeling | Retrieval cannot shorten the prompt | Cost per accepted review |
| Pricing data: How we verify → | |||
Use Gemini 3.1 Flash-Lite for compact, mechanically checked utility calls. Use Gemini 3.8 Flash when a user sees the answer or the request combines text with image, audio, or video inputs. A Pro model belongs behind an evaluation gate: stronger reasoning is useful only when it improves the completed task enough to cover its higher token and latency cost.
Gemini 3.1 Pro remains a preview. Pin an exact model ID; established routes and new releases can differ in compatibility and behavior.
Context, caching, audio, and grounding
Pro billing changes when the input prompt crosses the provider’s long-context threshold. The overview table shows base Standard rates, not every context tier. Use the Gemini 2.5 Pro and Gemini 3.1 Pro detail pages or the context cost guide to budget the higher input, output, and cached-input tiers. Retrieval, deduplication, and conversation summarization can keep many requests below the expensive tier without reducing answer quality.
Cached-token processing and cache storage are separate line items. A low cached-input rate does not make a long-lived cache free. Forecast the reusable prefix, expected hit rate, storage duration, invalidation frequency, and uncached remainder before claiming savings. Our cached tokens guide explains the break-even calculation.
Audio pricing can differ from text, image, or video input on some Gemini routes. Google Search and Maps grounding also add request- or query-based charges outside the token line. The calculator above estimates normalized token cost only, so a grounded or audio-heavy application needs those provider-specific charges added to its budget.
Free tier and hidden costs
Free access is model-, region-, and execution-class-specific. Treat it as an evaluation allowance, not a production budget. Confirm current eligibility, rate limits, data-use terms, and whether the exact model is available in the Gemini Developer API or Vertex AI route you plan to deploy.
The base token bill also omits several costs that commonly decide the real winner:
- retries after malformed structured output or tool failures
- grounding, retrieval, storage, and paid tool calls
- duplicated context and unused multimodal payloads
- fallback traffic sent to a second model or provider
- human review for customer-facing or regulated decisions
Track cost per accepted task rather than cost per call. A cheaper request is not cheaper if lower quality causes more retries, escalations, or manual correction.
A practical Gemini routing plan
Test short prompts, long documents, multimodal inputs, tools, and known failures. Compare Flash-Lite, Flash, and Argon or Pro with identical schemas, timeouts, retry caps, and response limits.
Record task success, input and output tokens, cache hits, grounding calls, latency, retries, safety refusals, and review time. Then deploy the least expensive route that clears the quality floor and escalate only uncertain or high-value requests. Recheck the token calculator with the measured input-to-output ratio instead of a generic average.
Include one independent provider in the test so the evaluation does not become a choice among Google tiers only. Compare Gemini with OpenAI pricing and Anthropic pricing. For a managed open-model control, check Novita’s current model catalog and confirm the exact model, region, and live rate before testing.
Affiliate disclosure: we may earn a commission from the sponsored link above. It does not affect our pricing data or recommendations.
Bottom line
Gemini 3.8 Flash is the sensible available production default to benchmark now, with Flash-Lite for validated utility work and Pro behind a quality gate. Evaluate Gemini 4 Argon now for difficult tasks, accounting for its December rate change and billed reasoning tokens. Include Flash’s January rate increase in longer-term budgets. The winning route is the one with the lowest cost per accepted task after context tiers, cache storage, grounding, retries, and review time are counted.
Current Standard billing, cache, grounding, Free-tier and scheduled-rate terms were reviewed against Google’s official Gemini API pricing. The source review record records this guide’s billing evidence; API access is not a subscription entitlement. Recheck the provider rate card before launching production traffic.
FAQ
Which Gemini API model is best for most developers?
Start with Gemini 3.8 Flash for user-facing and multimodal work. Use Gemini 3.1 Flash-Lite for simpler validated jobs and move to Pro only when your evaluation shows a worthwhile quality gain.
Will Gemini 3.8 Flash pricing change?
Google publishes introductory rates through December 31, 2026, and higher rates from January 1, 2027. The scheduled-rate table separates that future bill from today’s Standard token cost.
Does cached input always reduce the total Gemini bill?
No. Savings depend on eligible repeated input and actual cache hits, while cache storage and invalidation can add cost. Model both processing and storage before adopting a cache strategy.
Is grounding included in Gemini token pricing?
No. Search and Maps grounding can add separate request- or query-based charges, so the token calculator does not represent the entire grounded workload bill.
Does Gemini Pro have a free tier?
Availability varies by model and execution route. Check the current official row, limits, and terms before assuming preview, stable, Gemini Developer API, and Vertex AI access are interchangeable.