Gemini 3.8 Flash costs $0.75 per million input tokens and $3.75 per million output tokens—the same introductory rate Google set for 3.7 Flash. This rate runs through December 31, 2026. On January 1, 2027, it doubles to $1.50 and $7.50. The same increase applies to 3.6 Flash and 3.7 Flash. The per-token price did not change with this release.
What changed is how many tokens the model uses. Google says 3.8 Flash “works harder”: it takes more reasoning steps, calls tools iteratively, and “can use more tokens on longer running and complex tasks, by design.” Artificial Analysis measured the impact independently: its Intelligence Index cost $0.58 per task on 3.8 Flash at high thinking versus $0.40 on 3.7 Flash, with approximately 30% more output tokens per task.
Same sticker price, higher bill.
This guide covers the official Gemini API pricing page, explains the impact of each thinking level, and compares Gemini 3.8 Flash with Flash-Lite, Claude Sonnet 5, and GPT-5.6 Luna. For a model overview, start with what Gemini 3.8 Flash is.
Gemini 3.8 Flash pricing at a glance
All prices are per 1 million tokens on the paid tier.
| Rate | Intro (through Dec 31, 2026) | Standard (from Jan 1, 2027) |
|---|---|---|
| Input (text, image, video, audio, PDF) | $0.75 | $1.50 |
| Output (includes thinking tokens) | $3.75 | $7.50 |
| Context cache read | $0.075 | $0.15 |
| Cache storage (per 1M tokens per hour) | $0.50 | $1.00 |
| Batch API input (50% off) | $0.375 | $0.75 |
| Batch API output (50% off) | $1.875 | $3.75 |
| Google Search grounding | 5,000 free requests/month shared across Gemini 3.x, then $14 per 1,000 | Same |
| Free tier | $0, rate-limited, data used to improve Google products | Same |
Three details matter:
- Thinking tokens are billed as output at $3.75 per million, not as input.
- The introductory rate ends on December 31, 2026.
- 3.7 Flash and 3.6 Flash have the same January 1 price increase, so switching to an older Flash model does not avoid it.
The 3.7 Flash pricing breakdown covers the same rates for a model that uses fewer tokens per task.
Thinking tokens count as output tokens
Gemini 3.8 Flash reasons before answering, and every reasoning token is metered as output. A generateContent response reports:
-
usageMetadata.thoughtsTokenCount: internal reasoning tokens -
usageMetadata.candidatesTokenCount: visible answer tokens -
usageMetadata.promptTokenCount: input tokens
Both output counters use the same $3.75-per-million rate.
You control reasoning effort with thinking_level: low, medium, or high. On 3.8 Flash, the default is medium; Gemini 3 Pro defaults to high.
minimal is no longer supported and returns a validation error. Configurations migrated from earlier models should map minimal to low.
Google’s thinking documentation describes these settings as relative effort levels rather than fixed token budgets. Unlike the older integer thinking_budget, you cannot directly cap the number of thinking tokens. See the 3.8 Flash thinking levels guide for the latency and cost trade-offs.
Example request cost
Suppose a request uses 10,000 input tokens and returns 2,000 visible output tokens plus 6,000 thinking tokens.
At the introductory rate:
- Input:
10,000 × $0.75 / 1,000,000 = $0.0075 - Output:
(2,000 + 6,000) × $3.75 / 1,000,000 = $0.03 - Total:
$0.0375per request
Thinking accounts for 75% of the output charge and 60% of the total request cost.
From January 1, the same request costs:
- Input:
$0.015 - Output:
$0.06 - Total:
$0.075
“Same price as 3.7” is true but incomplete: the rate stayed the same while token usage increased.
Why cost per task increased
Google’s launch post describes the trade-off directly. On complex tasks, 3.8 Flash “executes extra reasoning steps, and calls tools iteratively,” verifying its work along the way.
More reasoning steps create more thinking tokens. Agent loops can also create additional tool-call turns.
Google’s recommended mitigations are to lower thinking_level or remain on 3.7 Flash.
Artificial Analysis measured the difference using nine evaluations in its Intelligence Index. Gemini 3.8 Flash at high scored 59 versus 56 for 3.7 Flash at high, while using approximately 48,000 output tokens per task—30% more than 3.7 Flash.
| Configuration | Cost per task (AA, intro rates) | Time per task |
|---|---|---|
| Gemini 3.8 Flash, high | $0.58 | 2.5 min |
| Gemini 3.8 Flash, medium | $0.41 | not published |
| Gemini 3.8 Flash, low | $0.24 | 0.8 min |
| Gemini 3.7 Flash, high | $0.40 | 2.2 min |
The results support two conclusions:
- 3.8 Flash at high costs 45% more per task than 3.7 Flash at high:
$0.58 / $0.40 = 1.45. - On 3.8 Flash, medium reduces cost by 29% compared with high:
$0.41 / $0.58 = 0.71. - Low reduces cost by 59% compared with high:
$0.24 / $0.58 = 0.41.
Medium on 3.8 Flash costs within one cent of high on 3.7 Flash, which is a useful benchmark when deciding whether to upgrade. The 3.8 Flash vs. 3.7 Flash comparison compares the models by workload.
Artificial Analysis also reports a blended price of $0.58 per million tokens at a 3:1 input-to-output ratio. Its matching the high-level cost per task is coincidental: the blended price is a per-token calculation, while task cost depends on how many tokens the model chooses to use.
Use the introductory rate before December 31
“Lock in” does not mean signing a contract. Google bills tokens at the rate in effect when they are consumed, so you cannot prepay 2027 usage at 2026 prices. Switching to 3.7 or 3.6 Flash also does not avoid the increase.
You can, however, move discretionary work into the introductory-rate window:
- Run backfills and one-off document processing through the Batch API. A 100-million-token backfill with 75 million input tokens and 25 million output tokens costs:
75 × $0.375 + 25 × $1.875 = $28.13 + $46.88 = $74.99
in batch during 2026. The same workload costs twice as much—$149.98—in January.
- Build your evaluation set now. Establish baseline token counts before the price change, so you can distinguish increased usage from increased rates.
-
Choose a thinking level per route. Every production route left on
mediumby default is a route whose cost has not been evaluated. Routes that pass your tests atlowshould uselowbefore January.
If price is the deciding factor, see the cheapest LLM API providers roundup. The Fable 5.1 and GPT-5.6 Sol comparison covers the premium tiers.
Gemini 3.8 Flash vs. Flash-Lite, Sonnet 5, and GPT-5.6 Luna
Per-token rates, per 1 million tokens:
| Model | Input | Output | Notes |
|---|---|---|---|
| Gemini 3.8 Flash (intro) | $0.75 | $3.75 | Thinking billed as output; cache read $0.075 |
| Gemini 3.8 Flash (from Jan 1) | $1.50 | $7.50 | Cache read $0.15 |
| Gemini 3.5 Flash-Lite | $0.30 | $2.50 | About 350 tok/s |
| Claude Sonnet 5 | $2 | $10 | Permanent; the Sep 1 rise was cancelled |
| GPT-5.6 Luna | $1 | $6 |
At introductory rates:
- 3.8 Flash input costs 2.5 times Flash-Lite’s input rate.
- 3.8 Flash output costs 1.5 times Flash-Lite’s output rate.
- 3.8 Flash is approximately 2.7 times cheaper than Sonnet 5 for both input (
$2 / $0.75) and output ($10 / $3.75). - 3.8 Flash is cheaper than Luna for input (
$0.75 vs. $1) and output ($3.75 vs. $6).
On January 1, the comparison changes:
- 3.8 Flash becomes more expensive than Luna for both input and output.
- The gap with Sonnet 5 narrows to approximately 1.3 times (
$2 / $1.50and$10 / $7.50). - Flash-Lite remains cheaper throughout.
If the introductory rate is why you selected 3.8 Flash, schedule a January re-evaluation now.
Per-token pricing is only half the calculation. A model with a higher rate can still cost less per task if it uses fewer tokens. Run the same task set through each candidate and compare the returned usage counts. The 3.8 Flash API guide shows how to read usageMetadata from both the Interactions API and legacy generateContent.
Catch cost regressions with Apidog token assertions
The main failure mode is not necessarily an incorrect answer. It is a correct answer that quietly uses 40% more tokens after a prompt or thinking-level change—until the invoice exposes the regression.
Apidog lets you treat token usage as an assertion target, just like an HTTP status code.
1. Store the API key securely
Create GEMINI_API_KEY in an Apidog environment and reference it as {{GEMINI_API_KEY}} in the x-goog-api-key header. The key never needs to be stored in the saved request.
2. Send a representative prompt
Save a POST request to:
https://generativelanguage.googleapis.com/v1beta/models/gemini-3.8-flash:generateContent
Use a representative production prompt and set an explicit thinkingConfig.thinkingLevel for each route.
3. Assert on usage fields
Add a post-processor script that reads usageMetadata and fails when usage exceeds your baseline:
const usage = pm.response.json().usageMetadata;
pm.test("thinking tokens within budget", () => {
pm.expect(usage.thoughtsTokenCount).to.be.below(8000);
});
pm.test("visible output within budget", () => {
pm.expect(usage.candidatesTokenCount).to.be.below(2500);
});
pm.test("request cost within budget", () => {
const cost = usage.promptTokenCount * 0.75 / 1e6
+ (usage.thoughtsTokenCount + usage.candidatesTokenCount) * 3.75 / 1e6;
pm.expect(cost).to.be.below(0.05);
});
4. Schedule the test
Put the requests in a test scenario and run it on a schedule. A token-usage increase then appears as a failed run the same day. The guide to scheduling API tests in Apidog covers the setup.
Update the two rate constants on January 1; the cost assertion will continue tracking the actual bill.
The same scenario can power a provider bake-off. Point copies at Flash-Lite, Sonnet 5, or Luna and compare usage fields side by side. Download Apidog to build the first scenario; the free plan supports this workflow.
FAQ
Does Gemini 3.8 Flash cost more than 3.7 Flash?
Per token, no. Both cost $0.75 per million input tokens and $3.75 per million output tokens through December 31, followed by $1.50 and $7.50 from January 1. Per task, yes: Artificial Analysis measured $0.58 per Intelligence Index task on 3.8 Flash at high versus $0.40 on 3.7 Flash because 3.8 Flash generates approximately 30% more output tokens.
Are thinking tokens billed separately?
No. They are billed as output tokens at $3.75 per million during the introductory period and appear in usageMetadata.thoughtsTokenCount. You control them indirectly through thinking_level; 3.8 Flash has no hard token budget, and minimal returns an error.
Is there a free tier for Gemini 3.8 Flash?
Yes. The AI Studio free tier charges nothing for input or output, subject to rate limits shown in AI Studio. Google uses free-tier data to improve its products.
The consumer Gemini app provides 3.8 Flash only to AI Pro and Ultra subscribers. See the free-usage guide for details.
What happens to costs on January 1, 2027?
Input, output, cache-read, cache-storage, and batch rates all double. If token usage per task remains constant, your bill doubles as well. The 3.6 and 3.7 Flash rates increase on the same date.
Which thinking level keeps costs down?
Start with Artificial Analysis’s measurements: $0.24 per task at low, $0.41 at medium, and $0.58 at high. Run your own evaluations at each level, keep the lowest setting that passes, and assert on token counts so the configuration cannot drift.
What to do this week
Gemini 3.8 Flash is priced like 3.7 Flash but uses tokens like a larger model.
This week:
- Set
thinking_levelper production route instead of accepting the default. - Move batch-friendly work into the introductory-rate window before December 31.
- Establish token baselines now.
- Add assertions for thinking tokens, visible output tokens, and estimated request cost.
- Schedule a provider and pricing re-evaluation for January 1, 2027.
That turns the January price increase into a planned variable instead of an invoice surprise.
Top comments (0)