Gemini 3.5 Flash Pricing: How Much Does It Actually Cost ?

Google shipped Gemini 3.5 Flash on May 19, 2026, with a bold pricing claim: “less than half the cost of other frontier models” for agentic tasks. This guide turns that claim into practical cost math you can use before wiring Flash into production.

Try Apidog today

You’ll get the per-token rates, free-tier caps, batch-mode discount, real-world workload estimates, and a side-by-side comparison against GPT-5.5 and Claude Opus 4.7. Use it to estimate your bill, choose the right tier, and identify where batch mode or caching can cut costs.

Quick summary

Cost type	Rate
Standard input	~$1.50 / 1M tokens
Standard output	~$9.00 / 1M tokens
Batch mode input	~$0.75 / 1M tokens (~50% off)
Batch mode output	~$4.50 / 1M tokens (~50% off)
Cached input	reduced rate (varies)
Free tier (AI Studio)	~1,500 requests/day, 1M tokens/min, 15 RPM
Vertex AI new account	$300 credit over 90 days

Rates are current as of May 2026 based on Google’s launch announcement and aggregator listings. Always verify against the official pricing page before committing budget.

Gemini 3.5 Flash per-token rates

Flash uses the same pay-as-you-go pattern as other Gemini models: input and output tokens are billed separately.

Tier	Input ($/1M)	Output ($/1M)
Standard	~$1.50	~$9.00
Cached input	discounted	n/a
Batch async	~$0.75	~$4.50

Two implementation details matter:

• Tokens are not words. A rough estimate is 1,000 tokens ≈ 750 English words. A 100,000-word document is about 133K input tokens.
• Output is much more expensive than input. Output costs roughly 6× more, so verbose responses are expensive. Prefer structured JSON or concise formats when possible.

Example response-shaping prompt:

Return only valid JSON matching this schema:

{
  "summary": "string, max 80 words",
  "risk_level": "low | medium | high",
  "action_items": ["string"]
}

For more background, see Gemini API batch mode is here and 50% cheaper.

Free tier: what you get without paying

The AI Studio free tier includes Flash from day one. Launch limits:

• 1,500 requests per day
• 1M tokens per minute
• 15 requests per minute

That is enough for side projects, internal prototypes, evaluation scripts, and small automation jobs.

Free-tier specifics:

• No credit card required
• Same gemini-3.5-flash model as the paid endpoint
• Same SDK pattern, using an AI Studio key
• Prompts may be used to improve Google’s models unless you opt out in AI Studio settings
• Quotas can change, so don’t build production capacity planning around the exact free-tier numbers

For setup walkthroughs, see How to use Gemini 3.5 Flash for free and How to get a free Google Gemini API key.

Batch mode: the 50% discount most teams miss

If your workload does not require real-time responses, batch mode cuts Flash costs roughly in half.

How it works:

1. Submit a batch job with up to 50,000 prompts.
2. Google processes them within 24 hours.
3. You pay about 50% less per token for both input and output.

Use batch mode for:

• Bulk document analysis
• Support ticket triage
• Content moderation
• Overnight content generation
• Historical data migration
• Backfills and reprocessing jobs

Avoid batch mode for:

• Chat UIs
• Real-time agents
• Interactive user workflows
• Anything where a user is waiting for a response

A simple routing rule:

if user_is_waiting:
  use standard Gemini 3.5 Flash
else:
  use batch mode

Most production systems should default to batch mode for any async workload. Setup details are in the batch mode guide.

Cached input: another cost lever

If your prompts share a long static prefix, context caching can reduce input cost.

Good cache candidates:

• System prompts
• Long instructions
• Policy documents
• Product documentation
• Reused reference material
• Retrieved chunks that appear across many queries

Pattern:

1. Cache a 100K-token reference document.
2. Reuse it across many queries.
3. Pay full rate only for the new user question and uncached context.

For RAG-style apps where repeated chunks appear across queries, expect the biggest savings when cache hit rates are high.

Real-world cost scenarios

Token math is easier with concrete examples. The following estimates use Flash standard rates.

Scenario 1: customer support chatbot

Assumptions:

• 10,000 user messages per day
• 200 input tokens per request
• 400 output tokens per response

Daily cost:

Input:
10,000 × 200 × ($1.50 / 1,000,000) = $3.00/day

Output:
10,000 × 400 × ($9.00 / 1,000,000) = $36.00/day

Total:
~$39/day
~$1,170/month

If responses can be batched, monthly cost drops to about $585/month. Add context caching for the system prompt and you can reduce input cost further.

Scenario 2: document Q&A SaaS

Assumptions:

• 1,000 documents analyzed per day
• 30K input tokens per document
• 500 output tokens per answer

Daily cost:

Input:
1,000 × 30,000 × ($1.50 / 1,000,000) = $45.00/day

Output:
1,000 × 500 × ($9.00 / 1,000,000) = $4.50/day

Total:
~$50/day
~$1,500/month

This is where Flash’s large context window is useful: you can send long documents directly instead of building complex chunking infrastructure for every workflow.

Scenario 3: long-running autonomous agent

Assumptions:

• One agent run has ~50 model turns
• Each turn averages 5K input tokens
• Each turn averages 1K output tokens
• 200 runs per day

Per-run cost:

Input:
50 × 5,000 × ($1.50 / 1,000,000) = $0.375

Output:
50 × 1,000 × ($9.00 / 1,000,000) = $0.45

Per run:
~$0.83

Daily total:

200 × $0.83 = ~$165/day
Monthly estimate = ~$4,950/month

For comparison, the same workload on Opus 4.7 at roughly $15/$75 per 1M tokens costs about $25/run, or about $5,000/day.

Scenario 4: chart extraction pipeline

Assumptions:

• 5,000 dashboard screenshots per day
• Each image input is equivalent to ~1,500 tokens
• Output is 300 tokens of structured JSON

Daily cost:

Input:
5,000 × 1,500 × ($1.50 / 1,000,000) = $11.25/day

Output:
5,000 × 300 × ($9.00 / 1,000,000) = $13.50/day

Total:
~$25/day
~$750/month

With batch mode, the same workload is about $375/month.

Scenario 5: high-volume content generation

Assumptions:

• 100,000 short articles per day
• 500 input tokens each
• 2,000 output tokens each

Daily cost:

Input:
100,000 × 500 × ($1.50 / 1,000,000) = $75/day

Output:
100,000 × 2,000 × ($9.00 / 1,000,000) = $1,800/day

Total:
~$1,875/day
~$56,250/month

Move this to batch mode and the monthly bill drops to about $28K. At this scale, also test routing simple work to cheaper models such as Gemini 3.1 Flash-Lite, while reserving Flash for harder generations.

Cost vs GPT-5.5 and Opus 4.7

Model	Input ($/1M)	Output ($/1M)	Multiple vs Flash
Gemini 3.5 Flash	~$1.50	~$9.00	1× baseline
GPT-5.5	~$10	~$30	6.7× input, 3.3× output
Claude Opus 4.7	~$15	~$75	10× input, 8.3× output

Run the customer support scenario through each model:

• Flash: $39/day
• GPT-5.5: ~$140/day
• Opus 4.7: ~$330/day

The flagship models may perform better on the hardest tasks, but for routine production workloads, Flash can be the lower-cost default.

For deeper comparisons, see GPT-5.5 pricing and the three-way comparison.

Cost vs other Gemini variants

Model	Input ($/1M)	Output ($/1M)	When to use
Gemini 3.1 Flash-Lite	~$0.40	~$2.00	High-volume routine work
Gemini 3 Flash	~$0.50	~$3.00	Last-generation, still solid
Gemini 3.1 Pro	~$2.00	~$12.00	Reasoning-heavy work pre-3.5 Pro
Gemini 3.5 Flash	~$1.50	~$9.00	New default for most workloads
Gemini 3.5 Pro June 2026	TBD	TBD	Hardest reasoning tasks

Flash is more expensive than older Flash variants, but cheaper than the previous Pro tier. For many teams, the practical routing strategy is:

routine/high-volume task  -> Flash-Lite
default production task   -> Gemini 3.5 Flash
hardest reasoning task    -> Pro or flagship model

For older Gemini pricing, see 3.1 Flash-Lite, 3.0 API pricing, and 3 Flash.

Vertex AI pricing for production

If you call Flash through Vertex AI instead of AI Studio, per-token pricing is the same. The main differences are operational:

• Service account auth instead of API keys
• Audit logs in Cloud Logging
• Data residency controls
• No free tier
• $300 new-account credit for 90 days
• Custom quotas available at scale

A practical adoption path:

1. Prototype with AI Studio free tier.
2. Move to AI Studio paid when you hit quota limits.
3. Move to Vertex AI when you need enterprise controls.

The model behavior is identical across these options.

Cost optimization checklist

Use these habits before scaling traffic:

1. Use batch mode for async workloads. Roughly 50% off with no quality loss.
2. Cache long static prefixes. Cache system prompts, reference docs, and reusable instructions.
3. Use structured output. JSON schemas reduce output length and make validation easier.
4. Route by task complexity. Send easy work to Flash-Lite, default work to Flash, and hard cases to Pro or flagship models.
5. Pre-validate inputs. Don’t spend tokens on malformed requests. Apidog can catch invalid request shapes before they hit the API.
6. Log token counts per prompt. Cost overruns usually come from a few long prompts or verbose outputs.

Example cost logging shape:

{
  "model": "gemini-3.5-flash",
  "input_tokens": 5200,
  "output_tokens": 430,
  "estimated_cost_usd": 0.01167,
  "route": "standard",
  "request_id": "req_123"
}

For prompt validation, download Apidog, save your Gemini endpoint as a request, and add response-shape assertions. This helps prevent repeated broken calls from burning quota during development.

When the free tier is not enough

Upgrade from free to paid Flash when:

1. You hit 1,500 requests/day repeatedly. The engineering time spent dodging quotas can cost more than pay-as-you-go usage.
2. You need higher RPM throughput. Free tier caps at 15 requests per minute.
3. You need audit logs or data residency. Move to Vertex AI on a billed account.

Many teams can replace free-tier juggling with a small paid Flash budget.

Pricing risks to watch

Three factors can change your cost model:

• Quota changes. Free-tier quotas can tighten over time.
• 3.5 Pro launch pricing. Flash pricing may shift when 3.5 Pro lands.
• Region surcharges. Vertex AI pricing varies by region; some regions may cost more.

Set cost alerts on day one. Use AI Studio quota controls or Cloud Billing budgets to cap daily spend.

Bottom line

Gemini 3.5 Flash is priced low enough to be the default starting point for many production AI workloads. Standard rates of roughly $1.50 input / $9 output per 1M tokens undercut other frontier-class options, and batch mode plus context caching can reduce effective cost further.

For tasks where Flash is not enough, route by complexity: use Flash for the bulk, and reserve models like GPT-5.5 or Opus 4.7 for the hardest prompts.

To implement this:

1. Save the Gemini 3.5 Flash endpoint in Apidog.
2. Build an eval set with 20 real prompts.
3. Compare Flash against your current model.
4. Log input and output tokens.
5. Extrapolate monthly cost.
6. Move eligible async workloads to batch mode.
7. Route simple tasks to cheaper models where quality holds.

That is usually enough work to identify meaningful savings within one billing cycle.