On this page · 10 sections
- What changes on 1 January 2027
- Who is affected: Gemini API versus Agent Platform
- Why your bill can rise by more than 2x
- Worked example: one month before and after
- Six ways to cut the bill
- Decision table: what to do by workload
- India and data residency notes
- Checklist before 1 January
- How eCorpIT can help
- References
Summary. On 1 January 2027 Google ends the introductory price for Gemini 3.8 Flash and Gemini 3.6 Flash. On the paid Gemini API, standard input goes from $0.75 to $1.50 per million tokens and output, which includes thinking tokens, from $3.75 to $7.50. Every other paid rate for these two models doubles on the same day: Batch and Flex move to $0.75 and $3.75, Priority to $2.70 and $13.50, cached input to $0.15 and cache storage to $1.00 per million tokens per hour. Google Cloud's Agent Platform takes the same step, with regional endpoints 10% dearer ($1.65 and $8.25 from January) and a 50% Provisioned Throughput credit that ends on 31 December 2026. Code that still calls gemini-3.7-flash or gemini-3.5-flash on the Gemini API has been routed to 3.8 Flash and 3.6 Flash since 8 October, so it is affected too. Gemini 3.5 Flash-Lite, at $0.30 and $2.50, has no dated change. In our worked example, a support assistant plus a nightly job goes from $6,075 a month today to $12,150 in January, and to $5,597 after Batch, caching, Flash-Lite routing, prompt trimming and lower thinking. A companion piece compares Gemini 3.8 Flash with Claude Haiku 5.5 and GPT-6 Luna.
About this article. This guide is for engineering leads, FinOps owners and CTOs who run Gemini 3.8 Flash or 3.6 Flash in production. We checked every price, date and model behaviour against Google's pricing pages, model documentation, release notes and deprecation schedule, Anthropic's and OpenAI's pricing pages, and Artificial Analysis benchmark data, on 1 November 2026. Vendors change prices often, so re-check before you commit; no vendor paid for inclusion. Researched with AI assistance and reviewed by our editorial team.
What changes on 1 January 2027
Google states the date in four places, and they agree. The Gemini API pricing page lists every paid Gemini 3.8 Flash and 3.6 Flash rate with one price through 31 December 2026 and another from 1 January 2027. The 3.8 Flash guide calls $0.75 and $3.75 an introductory rate and names $1.50 and $7.50 as standard pricing from 1 January. The 2 September 2026 launch post carries a footnote saying the introductory price expires on 31 December 2026, and the Agent Platform pricing page repeats the schedule for Google Cloud.
The Gemini API paid-tier rates, from Google's pricing page:
| Gemini Developer API, paid tier (per 1M tokens) | Through 31 Dec 2026 | From 1 Jan 2027 | Change |
|---|---|---|---|
| Standard input | $0.75 | $1.50 | 2x |
| Standard output, including thinking tokens | $3.75 | $7.50 | 2x |
| Cached input (Standard) | $0.075 | $0.15 | 2x |
| Cache storage, per 1M tokens per hour | $0.50 | $1.00 | 2x |
| Batch input / output | $0.375 / $1.875 | $0.75 / $3.75 | 2x |
| Flex input / output | $0.375 / $1.875 | $0.75 / $3.75 | 2x |
| Batch and Flex cached input | $0.0375 | $0.075 | 2x |
| Priority input / output | $1.35 / $6.75 | $2.70 / $13.50 | 2x |
| Grounding with Google Search | 5,000 free requests a month, then $14 per 1,000 | No dated change | None |
The figures are identical for gemini-3.8-flash and gemini-3.6-flash, so moving down a generation saves nothing per token. The free Search grounding allowance is shared across all Gemini 3 and newer models, and on the Gemini API it carries a default cap of 1,500 requests a day per project.
What else moves, and what does not
- Aliases. Since 8 October 2026, Gemini API requests to
gemini-3.7-flashare routed automatically togemini-3.8-flash, and requests togemini-3.5-flashgo togemini-3.6-flash. Neither endpoint is shutting down. The launch post had told efficiency-first users they could stay on 3.7 Flash, so anyone who did is now billed for 3.8 Flash.
- Text-to-speech. Gemini 3.8 Flash TTS audio output goes from $9.00 to $18.00 per million tokens and its text input from $0.50 to $1.00. Gemini 3.8 Flash-Lite TTS audio output goes from $6.00 to $12.00. Gemini 3.1 Flash TTS Preview has an earliest shutdown date of 17 November 2026, with the two 3.8 TTS models named as replacements, so TTS users meet the doubling either way.
- Managed agents. The Antigravity agent in Gemini managed agents, and the Antigravity SDK, use Gemini 3.8 Flash by default. Agent inference is billed at standard list rates.
- Gemini 4 Argon. Google's flagship doubles a month earlier, on 1 December 2026, from $2 and $10 to $4 and $20.
No dated change appears for Gemini 3.5 Flash-Lite ($0.30 input, $2.50 output; Batch and Flex $0.15 and $1.25), Gemini 3.1 Flash-Lite ($0.25 and $1.50), the Gemini 3.8 Live models, or Search and Maps grounding. The free tier still offers 3.8 Flash at no charge, but Google uses free-tier content to improve its products, which rules it out for production or personal data. The same trade applies to the invite-only Contributor tier, which lists 3.8 Flash at $0.375 and $1.875 with no end date.
Who is affected: Gemini API versus Agent Platform
The change is not limited to AI Studio keys. Google Cloud's Agent Platform pricing page, for the service many teams still call Vertex AI, puts Gemini 3.8, 3.7 and 3.6 Flash, and CodeMender running on them, on introductory pricing through 31 December 2026 and on $1.50 and $7.50 from 1 January 2027. A footnote on the same page says the promotional price is provided as 50% credits back on net spend, so check how your Cloud invoices show it before you reconcile January.
The table combines both price pages.
| Channel | Through 31 Dec 2026 (in / out) | From 1 Jan 2027 (in / out) | Notes |
|---|---|---|---|
| Gemini Developer API, paid tier | $0.75 / $3.75 | $1.50 / $7.50 | Batch, Flex, Priority and caching all double |
| Gemini Developer API, free tier | Free | Free | Content used to improve Google products |
| Gemini Developer API, Contributor tier | $0.375 / $1.875 | No dated change listed | Invite-only preview; content used to improve Google products |
| Agent Platform, global endpoint | $0.75 / $3.75 | $1.50 / $7.50 | Cached input $0.15 from January |
| Agent Platform, regional (non-global) endpoint | $0.825 / $4.125 | $1.65 / $8.25 | 10% regional premium; cached input $0.165 |
| Agent Platform Priority, global | $1.35 / $6.75 | $2.70 / $13.50 | Regional $2.97 / $14.85 from January |
| Agent Platform Flex and Batch, global | $0.375 / $1.875 | $0.75 / $3.75 | Regional $0.825 / $4.125 from January |
| Agent Platform Provisioned Throughput | 50% monthly billing credit on eligible spend | Credit window ends 31 Dec 2026 | Credit does not change configurations or burndown rates |
Two rows deserve attention. Teams on regional endpoints for data residency pay the 10% premium on top of the doubling. And the Provisioned Throughput credit is easy to miss: from 13 August to 31 December 2026, Google adds a monthly billing credit equal to 50% of net eligible Provisioned Throughput spend on Gemini 3.8, 3.7 and 3.6 Flash, issued on the 7th of the following month and valid for 30 days. Once the window closes, the same reserved capacity costs twice as much in effect.
If a gateway or FinOps dashboard calculates your spend from its own model price table, update it before 1 January, or January alerts will under-report by half. Our comparison of enterprise AI gateways covers where those tables live.
Why your bill can rise by more than 2x
The doubling is per token. Tokens per task are a separate variable, and on Gemini 3.8 Flash they have already gone up.
Google says so in its 3.8 Flash guide: the model can spend more tokens on long, complex tasks by design, because it reasons in smaller steps, calls tools repeatedly and checks its own work. Artificial Analysis, an independent benchmarking firm, measured the effect at launch. Average output tokens per task on its index rose 30% from Gemini 3.7 Flash to 48,000, and cost per task rose about 40%, from $0.40 to $0.58, at unchanged per-token prices.
Combine that with the alias routing above. A team whose gemini-3.7-flash calls now land on 3.8 Flash may already be spending more per task this quarter, and January doubles the rate on that larger token count. Check your own token logs before assuming the increase is exactly 100%.
The same launch data shows how much the thinking setting matters. At introductory prices, Artificial Analysis reported $0.58 per task at high thinking, $0.41 at medium (the default) and $0.24 at low, so low cost about 41% less than medium. Those are benchmark tasks rather than your traffic, but a gap that size is worth testing.
Worked example: one month before and after
The workload below is illustrative and uses only Google's list prices. Measure your real prompt sizes with our free LLM token counter before you rely on any of it.
A B2B SaaS company runs two Gemini 3.8 Flash workloads on the standard Gemini API tier:
- Support assistant: 500,000 requests a month. Each request sends 6,000 input tokens: a 4,500-token stable prefix (system prompt, policies, tool schemas) and 1,500 dynamic tokens (conversation and retrieved passages). Each returns 1,000 output tokens, including thinking at the default medium level.
- Nightly enrichment: 200,000 documents a month at 8,000 input and 1,000 output tokens each, currently sent through the standard API.
| Line item | Tokens per month | Through 31 Dec 2026 | From 1 Jan 2027, no changes |
|---|---|---|---|
| Assistant input | 3,000M | $2,250 | $4,500 |
| Assistant output | 500M | $1,875 | $3,750 |
| Enrichment input | 1,600M | $1,200 | $2,400 |
| Enrichment output | 200M | $750 | $1,500 |
| Total | $6,075 | $12,150 |
Now apply the levers one at a time, all at January prices:
| Step | Change | Monthly bill | Saving vs previous step |
|---|---|---|---|
| 0 | No changes | $12,150 | |
| 1 | Enrichment moves to the Batch API ($0.75 / $3.75) | $10,200 | $1,950 |
| 2 | Implicit caching hits 80% of the 4,500-token prefix | $7,770 | $2,430 |
| 3 | 40% of assistant turns (FAQ, classification) routed to Gemini 3.5 Flash-Lite | $6,302 | $1,468 |
| 4 | Dynamic context trimmed by 500 tokens per request | $6,047 | $255 |
| 5 | Low thinking on suitable 3.8 Flash turns: output falls from 1,000 to 800 tokens | $5,597 | $450 |
The arithmetic for each step:
- Step 1: 1,600M × $0.75 + 200M × $3.75 = $1,950, half the standard $3,900.
- Step 2: 3,600 cached tokens per request × 500,000 = 1,800M × $0.15 = $270. The remaining 1,200M input tokens × $1.50 = $1,800. Assistant input falls from $4,500 to $2,070.
- Step 3: 200,000 Flash-Lite requests: 1,200M × $0.30 + 200M × $2.50 = $860, with no caching assumed. The 300,000 requests left on 3.8 Flash: 1,080M cached × $0.15 + 720M × $1.50 + 300M × $7.50 = $162 + $1,080 + $2,250 = $3,492.
- Step 4: Flash-Lite input becomes 1,100M × $0.30 = $330, saving $30. On 3.8 Flash, uncached input becomes 570M × $1.50 = $855, saving $225.
- Step 5: 3.8 Flash output becomes 240M × $7.50 = $1,800, saving $450.
The end state, $5,597, is below today's $6,075, but only if the assumptions hold. An 80% cache hit rate needs steady traffic with a shared prefix, and Flash-Lite and lower thinking must pass your evals. On an Agent Platform regional endpoint, add 10% to every line. Explicit caches, which need the generateContent API, also bill storage: a 5,000-token cache held for a 720-hour month costs 0.005M × $1.00 × 720 = $3.60. These are calculations from published prices, not measurements of your traffic.
Six ways to cut the bill
1. Make caching actually hit
Google's context caching guide says implicit caching is on by default for Gemini 2.5 and newer models, with nothing to configure. The catch is the minimum: a request needs at least 4,096 input tokens on Gemini 3.8 Flash or 3.6 Flash before a cache hit is possible, so short prompts earn nothing. Consolidate stable material (instructions, policies, tool schemas, few-shot examples) into one prefix at the very start of the prompt, ideally large enough to clear that threshold on its own.
Send requests that share a prefix close together in time, and read usage.total_cached_tokens in the response to confirm hits. The Interactions API supports implicit caching only; explicit caches need the generateContent API.
2. Move anything that can wait to Batch or Flex
Both tiers cost 50% of standard, which brings 3.8 Flash back to $0.75 and $3.75 in January. Batch is asynchronous, with up to 24 hours of turnaround. Flex, still in preview, is synchronous but best-effort: Google's Flex guide targets 1 to 15 minutes, may shed requests with 503 errors, never upgrades a request to Standard on its own, and recommends client timeouts of 10 minutes or more. It suits offline evals, background agents and chains where each step depends on the last.
from google import genai
client = genai.Client(http_options={"timeout": 900000}) # 15 minutes, in milliseconds
interaction = client.interactions.create(
model="gemini-3.8-flash",
input="Classify and summarise this support ticket...",
service_tier="flex", # 50% of Standard; add your own retry with backoff
)
print(interaction.output_text)
3. Turn thinking down where quality allows
Gemini 3.8 Flash supports low, medium (the default) and high thinking levels; minimal returns an error. Thinking tokens bill at the output rate, the line that reaches $7.50 in January. Google recommends low effort for latency-critical work such as real-time chat and drafting. At launch, Artificial Analysis scored low at 52 on its index against 57 at medium. Those scores come from the index version in use before Artificial Analysis moved to v5 on 27 October 2026, so treat them as relative. Run your evals at both levels and keep low wherever scores hold.
4. Route simple traffic to Flash-Lite, not to 3.6 Flash
Because 3.6 Flash doubles in step with 3.8 Flash, it is not a cheaper fallback. The cheaper Google option is Gemini 3.5 Flash-Lite at $0.30 input and $2.50 output, with no dated change. From January that is 80% below 3.8 Flash on input and 67% below on output. Google positions it as a cost-efficient model for high-volume agentic tasks, translation and simple data processing, which covers most classification, routing, extraction and FAQ traffic.
Avoid three tempting shortcuts. Gemini 3.1 Flash-Lite is cheaper at $0.25 and $1.50, but its earliest shutdown date is 7 May 2027 and Google names 3.5 Flash-Lite as its replacement. Google limits Gemini 2.5 Flash to users who have used it before. Gemini 3 Flash Preview, at $0.50 and $3.00, is a preview model for which Google already names 3.6 Flash as the replacement. Our model routing framework explains how to decide which turns are safe to move.
5. Trim prompts and cap output
Every token you stop sending is worth twice as much in January. The usual waste is retrieval that returns too many passages, full conversation history where a summary would do, and schemas for tools the current step cannot call. Set a maximum output token limit, and use structured outputs where you need fields rather than prose. Trace the effect per workload with an LLM observability tool.
6. Price the alternatives before you renew anything
Claude Haiku 5.5 lists at $0.10 input and $0.50 output per million tokens for prompts up to 100,000 tokens, and GPT-6 Luna at the same rates for prompts up to 272,000 input tokens. From January that is one fifteenth of Gemini 3.8 Flash's standard rate. Per-token price is not per-task cost, and the models differ in input types, long-prompt rules and benchmark results, so read our Gemini 3.8 Flash vs Claude Haiku 5.5 vs GPT-6 Luna comparison before moving production traffic.
Decision table: what to do by workload
| If your workload is | Do this before 1 January |
|---|---|
| Interactive chat or agents with long system prompts | Build a stable prefix, place it first, keep requests above 4,096 tokens and confirm cached-token counts |
| Nightly or offline jobs (enrichment, evals, backfills) | Move to the Batch API; use Flex for synchronous chains |
| High-volume classification, routing or extraction | Test Gemini 3.5 Flash-Lite against your eval set |
| Long-horizon coding agents or Antigravity managed agents | Keep 3.8 Flash at medium, set per-task token caps and budget for 2x |
| Agent Platform regional endpoints for residency | Budget $1.65 / $8.25 and confirm whether a global endpoint meets your data policy |
| Provisioned Throughput on Agent Platform | Re-model reserved capacity against pay-as-you-go now the 50% credit ends |
| Text-to-speech on Gemini 3.8 Flash TTS | Budget 2x; 3.8 Flash-Lite TTS also doubles but stays a third cheaper on audio output |
| Gemini spend under about $500 a month | Caching and Batch only; routing rarely pays back the engineering time at this size |
India and data residency notes
For teams serving Indian users, two practical points sit outside the price table.
First, residency costs more. On Agent Platform, non-global endpoints carry a 10% premium, so in-region processing moves from $0.825 to $1.65 per million input tokens in January. We found no Google page confirming a Mumbai (asia-south1) API endpoint for Gemini 3.8 Flash as of 1 November 2026. Google's data residency page for the Gemini Enterprise app says its India in-country region does not offer 3.8 Flash, and that traffic from unsupported in-country regions goes to the global endpoint once an administrator accepts a warning. That page covers the app rather than the API, but it points the same way. A 21 September 2026 post on Google's developer forum, from an Indian team evaluating Gemini 2.5, 3.1 and 3.5 Flash-Lite for a voice agent, reported seeing only global and us-central1 options. Check the current Agent Platform model locations before you promise in-country processing.
Second, keep personal data off free keys. India's Digital Personal Data Protection Act 2023 makes the data fiduciary responsible for processing done on its behalf by a data processor, and allows a processor to be engaged only under a valid contract. Google states that free-tier content is used to improve its products and paid-tier content is not. Note too that since 26 October 2026, new paid-tier projects using the generateContent API store requests by default; opt out per request or in the logging settings if your policy requires it. Our DPDP engineering playbook covers the wider obligations. EU teams using regional endpoints for GDPR reasons pay the same 10% regional premium.
Checklist before 1 January
- Search code and config for
gemini-3.8-flash,gemini-3.7-flash,gemini-3.6-flashandgemini-3.5-flash, including agent configs.
- Export November and December usage split into input, cached input, output and thinking tokens, per workload.
- Update price tables in your gateway, FinOps dashboards and any customer-facing usage pricing.
- Put a stable prefix first, keep cache-eligible requests above 4,096 tokens, then verify cache hits.
- Move offline jobs to Batch, and latency-tolerant chains to Flex with retries and long timeouts.
- Run evals on 3.8 Flash at low and medium thinking, and on 3.5 Flash-Lite for your simplest intents.
- Reset Google Cloud budgets and alerts to January rates, and re-model any Provisioned Throughput without the credit.
- In the first week of January, compare the first invoice lines with your forecast.
How eCorpIT can help
eCorpIT is a Gurugram-based technology organisation, founded in 2021, assessed at CMMI Level 5 and MSME certified, with senior-led engineering teams working across AWS, Microsoft and Google platforms. Our LLM migration and cost optimisation service measures token use per workload, restructures prompts for caching, moves offline jobs to Batch or Flex and runs evals on your own traffic before any routing or vendor change goes live. Talk to us at /contact-us/.
Last updated: 1 November 2026.
