Gemini 3.8 Flash API Pricing: $0.75 Now, $1.50 in January

Gemini 3.8 Flash costs $0.75 per million input tokens and $3.75 per million output tokens, including thinking tokens, until 31 December 2026. On 1 January 2027 both numbers double. The model ID is gemini-3.8-flash, it is GA on the Gemini API and on Vertex AI, and it takes three thinking levels. Everything below is read from Google’s pricing and rate-limit pages on 3 September 2026, the day after launch.

Model ID:     gemini-3.8-flash
Standard:     $0.75 in / $3.75 out per 1M through 2026-12-31
              $1.50 / $7.50 from 2027-01-01
Batch, Flex:  $0.375 / $1.875, then $0.75 / $3.75
Priority:     $1.35 / $6.75, then $2.70 / $13.50
Cache:        $0.075 read (then $0.15); $0.50 per 1M per hour storage (then $1.00)
Free tier:    yes, "Free of charge" column on the pricing page
Context:      1,048,576 in / 65,536 out
Thinking:     low | medium (default) | high; minimal returns an error
Vertex AI:    GA, 2 Sep 2026, global + us / eu multi-region
Gateways:     not in the Ofox catalog as of 2026-09-03

Last updated 3 September 2026. The introductory price expires on 31 December 2026; budgets that run past that date need the second row.

How Much Does the Gemini 3.8 Flash API Cost?

$0.75 in, $3.75 out, and the pricing page prints the 2027 rate in the same cell. Google’s pricing table for Gemini 3.8 Flash reads, verbatim, “$0.75 through December 31, 2026. $1.50 starting January 1, 2027.” for input and “$3.75 through December 31, 2026. $7.50 starting January 1, 2027.” for output.

Three details in that table decide more of the bill than the headline rate:

• Output price includes thinking tokens. Reasoning is metered at $3.75 per million. The effort level is a price multiplier.
• Batch and Flex are exactly half. If a workload can wait, the same model costs $0.375 / $1.875.
• Priority is 1.8x. $1.35 / $6.75 buys a separate rate-limit pool, not a different model.

Gemini 3.7 Flash and 3.6 Flash carry identical rows. There is no per-token saving in moving between them, which the 3.8 Flash vs 3.7 Flash comparison covers in detail.

Is Gemini 3.8 Flash Free?

Yes, on the free tier, with limits Google does not publish. The pricing page has a free-tier column for Gemini 3.8 Flash and it reads “Free of charge” for input, output and context caching. Grounding is 5,000 free requests a month at every tier.

What the free tier does not come with is a public number for requests per minute. The rate-limits page says rate limits “can be viewed in Google AI Studio” and that “Specified rate limits are not guaranteed and actual capacity may vary.” Third-party guides quote figures such as 10 RPM and 1,500 requests per day for the free tier; treat those as unverified, and read the AI Studio rate-limit page for your own project instead.

What Are the Gemini 3.8 Flash Rate Limits?

Per-model RPM and TPM are shown only in AI Studio. What Google publishes is the tier ladder, the spend-based ceiling and the batch ceilings.

All figures from Google’s rate-limits page on 3 September 2026. Four rules from the same page matter in production:

• Limits are per project, not per API key. Rotating keys inside one project buys nothing. Daily quotas reset at midnight Pacific.
• The spend-based limit returns 429 RESOURCE_EXHAUSTED, the same code as an RPM limit. At Tier 1, $10 in 10 minutes is roughly 2.7M output tokens at the standard rate; a burst of long high-effort calls can hit it well before any request-count limit.
• Priority inference has its own pool at 0.3x the standard rate limit for the model and tier.
• Tier upgrades are automatic once the spend and age criteria are met: instant from Free to Tier 1, within 10 minutes after that.

If you are already seeing 429s on Gemini calls, Google’s own error code reference lists rate_limit_exceeded against quota_exceeded as the two 429 causes to tell apart, and the Claude Code 429 guide covers the same header-reading and backoff mechanics that apply to any token-bucket API, Gemini included.

What You Actually Pay per Task

Same price per token as 3.7 Flash, about 45% more per task at high effort, because the model writes more. Google’s launch post says so directly: “At times, the model might use more tokens to maximize performance, especially at higher effort levels.” The developer docs add: “Gemini 3.8 Flash can use more tokens on longer running and complex tasks, by design.”

Artificial Analysis put numbers on it:

Running the full index cost AA $825.83 on 3.8 Flash against $484.73 on 3.7 Flash. The default medium level is the pricing story here: 57 points for $0.41 per task is one point above 3.7 Flash at high for about a cent more. Google’s HLE-Verified score for the model is 54.9%, from the same launch post.

A budget built on 3.7 Flash’s output volume will be wrong on 3.8 Flash at high effort. Read completion_tokens (OpenAI-compatible) or usageMetadata.candidatesTokenCount (native) on your own traffic before committing.

How to Get a Gemini 3.8 Flash API Key

Go to Google AI Studio and click “Create API key.” No separate signup. Google’s API key guide states: “If you are a new user, Google AI Studio automatically creates a default Google Cloud project and API key after you accept the Terms of Service.” An existing Google Cloud user picks a project from the dashboard instead. The key works immediately on the free tier; no card is required to start.

Two things to do before shipping it:

• Never put the key in client-side code. The same guide is explicit: “Keys compiled in client-side code can be extracted by users.” Route browser or mobile traffic through your own backend, or through a gateway that issues its own scoped key.
• Migrate off Standard keys before the deadline. Google’s page states: “On September 2026: the Gemini API will reject requests from Standard keys. You must migrate to auth keys before this date to avoid service interruption.” Standard keys bill to a project but don’t identify the caller; auth keys bind to a service account and get faster leaked-key enforcement. The guide gives no day within the month, only “September 2026” — for a key created today, that migration is not optional, it is due this month.

How to Access Gemini 3.8 Flash

Three doors: the Gemini API with an AI Studio key, Vertex AI, or an OpenAI-compatible gateway once it lists the model.

Gemini API, native protocol

curl "https://generativelanguage.googleapis.com/v1beta/models/gemini-3.8-flash:generateContent" \
  -H "x-goog-api-key: $GEMINI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [{"parts": [{"text": "Rewrite this GROUP BY as a window function."}]}],
    "generationConfig": {"thinkingConfig": {"thinkingLevel": "medium"}}
  }'

thinkingLevel accepts low, medium and high. minimal returns a validation error; the 3.7 Flash guide shows the exact 400 body that produces, and 3.8 Flash has the same restriction.

Gemini API, OpenAI-compatible

from openai import OpenAI

client = OpenAI(
    base_url="https://generativelanguage.googleapis.com/v1beta/openai/",
    api_key="YOUR_GEMINI_API_KEY",
)
r = client.chat.completions.create(
    model="gemini-3.8-flash",
    reasoning_effort="medium",
    messages=[{"role": "user", "content": "Rewrite this GROUP BY as a window function."}],
)
print(r.usage.completion_tokens)

On this surface the parameter is reasoning_effort, mapped to the same three levels.

Vertex AI

Google’s Vertex model page lists gemini-3.8-flash as “Launch stage: GA, Release date: September 2, 2026”, available on the global endpoint and the us and eu multi-regions, with Standard, Flex and Priority pay-as-you-go plus Provisioned Throughput. Vertex bills through Google Cloud and has its own quota model; Vertex is also Google’s suggested route if the Gemini API rejects your location.

Through Ofox

Gemini 3.8 Flash is not in the Ofox catalog as of 3 September 2026. google/gemini-3.7-flash is, at $0.75 input and $3.75 output on its Google provider route (the CloudVertex route on the same model page is $1.50 / $7.50, and that higher figure is what the pricing field in /v1/models currently reports), reachable over the OpenAI-compatible endpoint at https://api.ofox.ai/v1 and over Google’s native Gemini protocol at https://api.ofox.ai/gemini. When 3.8 Flash is added the call is the same with the google/ prefix on the model string. The live catalog at GET https://api.ofox.ai/v1/models is the authority on what you can call; do not assume from this page.

client = OpenAI(base_url="https://api.ofox.ai/v1", api_key="YOUR_OFOX_API_KEY")
r = client.chat.completions.create(
    model="google/gemini-3.7-flash",   # swap to google/gemini-3.8-flash once listed
    reasoning_effort="medium",
    messages=[{"role": "user", "content": "ping"}],
)

Gemini 3.8 Flash Pricing vs the Alternatives

Gemini prices from Google’s pricing page on 3 September 2026, DeepSeek from its API pricing page; index scores from Artificial Analysis the same day. The cheapest current-generation Flash by per-token rate is 3.8 Flash tied with 3.7 and 3.6. Per task, on AA’s measurement, 3.8 Flash at low is cheapest at $0.24 but scores 52; at the 56 to 57 level, 3.7 Flash at high and 3.8 Flash at medium are a cent apart.

The Date to Put in Your Calendar

On 1 January 2027 every Gemini 3.8 Flash row doubles: $1.50 / $7.50 standard, $0.75 / $3.75 Batch and Flex, $2.70 / $13.50 Priority, $0.15 cache read, $1.00 per million per hour storage. If token counts hold, AA’s $0.58 / $0.41 / $0.24 per-task ladder becomes roughly $1.16 / $0.82 / $0.48. Google has printed this on the pricing page since launch, so it is not a rumour to hedge; it is a line item to schedule.

Sources

• https://ai.google.dev/gemini-api/docs/pricing
• https://ai.google.dev/gemini-api/docs/rate-limits
• https://ai.google.dev/gemini-api/docs/latest-model
• https://blog.google/innovation-and-ai/models-and-research/gemini-models/3-8-flash-and-3-8-flash-cyber/
• https://docs.cloud.google.com/vertex-ai/generative-ai/docs/models/gemini/3-8-flash
• https://artificialanalysis.ai/articles/gemini-3-8-flash
• https://ofox.ai/models/google/gemini-3.7-flash

Prices, tier rules and batch ceilings read from Google’s pages on 3 September 2026. Ofox catalog checked against the live /v1/models endpoint the same day.