As of 14 November 2026, Gemini 4 Carbon and Gemini 4 Argon cost $2 per 1M input tokens and $10 per 1M output tokens until 30 November, then $4 and $20 from 1 December. Gemini 3.5 Flash-Lite costs $0.30 and $2.50, and Gemini 3.1 Pro Preview costs $2 and $12 for prompts up to 200,000 tokens.
Those are Google's US-dollar list prices on its Gemini Developer API pricing page, read on 14 November 2026. Gemini 4 Hydrogen, the Flash models and the Flash-Lite models also run free on the free tier. The catch there is your data, not your wallet.
On this page
- TL;DR
- How I read Google's price list
- Prices per 1M tokens, model by model
- Intro prices end 1 Dec and 1 Jan
- Is the Gemini API free?
- Thinking tokens bill as output
- Batch, Flex and Priority
- Context caching
- Grounding with Google Search
- Three monthly bills
- How billing works
- Images, video, voice, embeddings
- Agent Platform (old Vertex AI)
- vs OpenAI, Claude and DeepSeek
- Which model to pay for
- FAQ
- Sources
Four extras decide the real bill: thinking tokens, context caching, grounding with Google Search and the Batch discount. Each gets a section below, and three worked monthly bills put them together. If you pay for the Gemini app rather than the API, that is a different price list, and my guide to Google's Gemini plans covers it.
Gemini API pricing in five lines (November 2026)
- Per 1M tokens, read 14 November 2026: Gemini 4 Carbon and Argon cost $2 input and $10 output until 30 November, then $4 and $20; Hydrogen $1.50 and $7.50; 3.8 Flash $0.75 and $3.75 until 31 December, then $1.50 and $7.50; 3.5 Flash-Lite $0.30 and $2.50.
- Thinking tokens bill as output. Batch and Flex are 50% off Standard on Google's tier table; Priority costs 1.8 times Standard on Google's tables.
- The free tier covers Hydrogen, the Flash and Flash-Lite models, within limits Google shows only in AI Studio; Carbon, Argon and 3.1 Pro Preview are paid only, and Google may use free-tier prompts to improve its products.
- "Gemini 3 Pro" was shut down on 9 March 2026; its name now points to 3.1 Pro Preview ($2 and $12 up to 200k-token prompts). 2.5 Pro and 2.5 Flash calls reroute on 1 December.
- Billing is prepaid from $5: credits expire after 12 months and are not refundable, and a new paid account is capped at $250 a month until it has paid $100 and waited three days.
How I read Google's price list
I read Google's Gemini API pricing, models, release notes, deprecations, rate-limits, billing and terms pages, Google Cloud's Agent Platform pricing page, and the API pricing pages of OpenAI, Anthropic and DeepSeek on 14 November 2026. I opened no Google account, created no API key and sent no request. Every monthly bill below is my arithmetic on list prices, and I label it that way.
Google moved its docs. Every ai.google.dev/gemini-api/docs address now redirects to aistudio.google.com/docs, under a banner that reads "Welcome to the new home for Gemini API documentation in Google AI Studio." The pricing table prints its price changes as dated rows inside each cell, so the intro price and the price that replaces it sit side by side. Loaded from India, the page showed the same dollar figures as the US read. No other currency appeared.
Older guides still price Gemini 3 Pro, 2.5 Pro and a "3.7 Flash", and some call Hydrogen a Flash model. Google has shut down 3 Pro, routed 3.7 Flash into 3.8 Flash and deprecated 2.5 Pro, and it never calls Hydrogen a Flash model. Every Gemini figure in this post is Google's page as it stood on 14 November 2026.
Gemini API prices per 1M tokens, model by model
Every text model on Google's pricing page, Standard tier, in US dollars per 1M tokens as of 14 November 2026. Where a cell holds two prices, the first runs to the date shown and the second replaces it.
| Model | Status | Free tier | Input | Output (incl. thinking) | Cached input | Batch (in / out) |
|---|---|---|---|---|---|---|
| Gemini 4 Carbon | GA, 10 Nov 2026 | No | $2.00 to 30 Nov; $4.00 from 1 Dec | $10.00 to 30 Nov; $20.00 from 1 Dec | $0.10; $0.20 from 1 Dec | $1 / $5; $2 / $10 from 1 Dec |
| Gemini 4 Argon | GA, 13 Oct 2026 | No | $2.00 to 30 Nov; $4.00 from 1 Dec | $10.00 to 30 Nov; $20.00 from 1 Dec | $0.10; $0.20 from 1 Dec | $1 / $5; $2 / $10 from 1 Dec |
| Gemini 4 Hydrogen | GA, 12 Nov 2026 | Yes | $1.50 | $7.50 | $0.075 | $0.75 / $3.75 |
| Gemini 3.8 Flash | GA, 2 Sep 2026 | Yes | $0.75 to 31 Dec; $1.50 from 1 Jan 2027 | $3.75 to 31 Dec; $7.50 from 1 Jan 2027 | $0.075; $0.15 from 1 Jan | $0.375 / $1.875; $0.75 / $3.75 from 1 Jan |
| Gemini 3.6 Flash | GA, 21 Jul 2026 | Yes | $0.75 to 31 Dec; $1.50 from 1 Jan 2027 | $3.75 to 31 Dec; $7.50 from 1 Jan 2027 | $0.075; $0.15 from 1 Jan | $0.375 / $1.875; $0.75 / $3.75 from 1 Jan |
| Gemini 3.5 Flash-Lite | GA, 21 Jul 2026 | Yes | $0.30 | $2.50 | $0.03 | $0.15 / $1.25 |
| Gemini 3.1 Flash-Lite | GA; earliest shutdown 7 May 2027 | Yes | $0.25 ($0.50 audio) | $1.50 | $0.025 | $0.125 / $0.75 |
| Gemini 3.1 Pro Preview | Preview since 19 Feb 2026 | No | $2.00 up to 200k; $4.00 above | $12.00 up to 200k; $18.00 above | $0.20; $0.40 above 200k | $1 / $6 up to 200k; $2 / $9 above |
| Gemini 3 Flash Preview | Preview, "Our legacy Flash model" | Yes | $0.50 ($1.00 audio) | $3.00 | Not recorded | $0.25 / $1.50 |
Google's Gemini Developer API pricing page, Standard and Batch tabs, read 14 November 2026. Carbon and Argon change on 1 December 2026, 3.8 and 3.6 Flash on 1 January 2027. "Not recorded" means my read did not capture that cell. Gemini 2.5 Flash-Lite ($0.10 / $0.40) is left out: Google now limits the 2.5 models to people who already used them.
Read the Status column before the price column. The cheapest row, 3.1 Flash-Lite, has an earliest shutdown date of 7 May 2027, and Google names 3.5 Flash-Lite as its replacement. Google's models page says which two to start on: "For any new projects, use our latest models: Gemini 4 Hydrogen or 3.5 Flash-Lite."

Carbon and Argon cost exactly the same, so pick between them on Google's speed claim.
Google gives the two identical prices in every tab: Standard, Batch, Flex and Priority. Its Carbon guide says both have a 1M-token context window, long decode up to 1M tokens and tunable thinking levels "at the same per-token price." Neither is on the free tier.
The descriptions differ. Carbon is "Our most capable reasoning model". Argon is "Our efficient Gemini 4 frontier reasoning model", which Google says gives "faster responses and lower token spend per turn compared to Gemini 4 Carbon". That claim is vendor-reported and comes with no figure. If it holds on your prompts, Argon finishes a job on fewer billed tokens, and the only way to find out is to run the same prompts through both and compare the usage counts.
Only 3.1 Pro Preview still charges more above 200,000 prompt tokens.
Gemini 3.1 Pro Preview costs $2 input and $12 output per 1M tokens up to 200,000 prompt tokens, and $4 and $18 above that, as of November 2026. Google Cloud's Agent Platform page spells out the rule: "If a query input context is longer than 200K tokens, all tokens (input and output) are charged at long context rates." Gemini 2.5 Pro has a tier like it, and its requests reroute to 3.8 Flash on 1 December. The Gemini 4 models charge one price across the whole window; Google's Hydrogen guide says it "uses flat standard pricing on the paid tier across the full 1M context window".
Introductory prices end on 1 December and 1 January
Two price steps land within seven weeks of this post. Both are printed on Google's pricing page, read 14 November 2026, as dated rows:
| Model | Price per 1M (in / out) | Until | Then | From |
|---|---|---|---|---|
| Gemini 4 Carbon and Argon | $2 / $10 | 30 Nov 2026 | $4 / $20 | 1 Dec 2026 |
| Gemini 3.8 Flash and 3.6 Flash | $0.75 / $3.75 | 31 Dec 2026 | $1.50 / $7.50 | 1 Jan 2027 |
| Gemini 3.8 Flash TTS | $0.50 text in / $9.00 audio out | 31 Dec 2026 | $1.00 / $18.00 | 1 Jan 2027 |
| Gemini 3.8 Flash-Lite TTS | $0.50 / $6.00 | 31 Dec 2026 | $1.00 / $12.00 | 1 Jan 2027 |
Google's Gemini Developer API pricing page, read 14 November 2026. Cached input and cache storage on 3.8 and 3.6 Flash double on the same date.
Take a coding agent that uses 50M input and 10M output tokens a month on Carbon. It pays $200 in November and $400 from December, by my arithmetic. Same code, same prompts, double the bill. Google's terms allow it: price changes take effect "30 days after they are posted unless otherwise specified". So price any long-lived project on the second number.
Routing changes the bill without a code change too. Google's deprecation and release notes set these moves:
- From 1 December 2026, every request to
gemini-2.5-pro"will automatically route to Gemini 3.8 Flash (gemini-3.8-flash) and be billed at Gemini 3.8 Flash rates." 2.5 Pro costs $1.25 / $10 up to 200k prompt tokens until then, so December's rate is lower, and January's step doubles it. - From the same day,
gemini-2.5-flashrequests run on 3.5 Flash-Lite at $0.30 / $2.50, the per-token rate 2.5 Flash charges today. - 3.7 Flash and 3.5 Flash were deprecated on 8 October 2026; their requests already go to 3.8 Flash and 3.6 Flash.
- Omni 1.1 Flash video moves to Omni 1.2 Flash on 3 December.
For the longer record, my tracker shows how API prices have moved since 2023.

"Gemini 3 Pro" now means 3.1 Pro Preview.
Google released gemini-3-pro-preview on 18 November 2025 and shut it down on 9 March 2026. The release note that day reads: "The Gemini 3 Pro Preview model has been shut down. The gemini-3-pro-preview now points to gemini-3.1-pro-preview."
So a Gemini 3 Pro price today is the 3.1 Pro Preview price: $2 input and $12 output per 1M tokens for prompts up to 200,000 tokens, as of November 2026. 3.1 Pro Preview has no announced shutdown date, but it is still a preview, and Google's models page says previews "will be deprecated with at least 2 weeks notice." Keep a fallback model wired in if you build on it.
Is the Gemini API free? On some models, and Google can learn from it
Yes, for part of the lineup. Google's pricing page puts the trade on its own cards: the Free card says "Content used to improve our products", and the Paid card says "Content not used to improve our products". As of 14 November 2026:
- Free: Gemini 4 Hydrogen, 3.8 Flash, 3.6 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite and 3 Flash Preview, plus the TTS, Live and Transcribe models, both embedding models and Gemma 4, which has no paid tier at all.
- Paid only: Carbon, Argon and 3.1 Pro Preview, plus every image and video model, cost money from the first token.
- Grounding with Google Search on Gemini 3 and newer is not available on the free tier, though Google says you can test it in AI Studio.
Google no longer prints per-model free limits in its docs. The rate-limits page says limits "depend on a variety of factors (such as your usage tier) and can be viewed in Google AI Studio", and that "Specified rate limits are not guaranteed and actual capacity may vary." They apply per project, not per key, and daily counts reset at midnight Pacific time. Any per-model number in an older guide is one I could not verify, so I give none.
The data terms are the real price. Google's Gemini API Additional Terms, effective 23 March 2026, say that on unpaid services Google uses what you send and what comes back "to provide, improve, and develop Google products and services and machine learning technologies". Two more lines from the same terms:
- "To help with quality and improve our products, human reviewers may read, annotate, and process your API input and output."
- "Do not submit sensitive, confidential, or personal information to the Unpaid Services."
Fine for a demo. Not for customer data. On the paid tier, Google says it does not use your prompts, files or responses to improve its products, but it still "logs prompts and responses for a limited period of time, solely for detecting and preventing violations of the Prohibited Use Policy". Since 26 October 2026, new paid projects on the Generate Content API also "now store requests by default (store=true)"; set store=false, or change it in Logs and Datasets settings, to opt out.
If you build for Europe or the UK, one rule decides it: "You may use only Paid Services when making API Clients available to users in the European Economic Area, Switzerland, or the United Kingdom." For people in those regions, the paid data terms cover the free quota too. These are Google's words, not legal advice. My tracker of free LLM API tiers shows how Gemini's free tier compares with other providers'.

Thinking tokens are billed as output
Every output price on Google's table carries the label "Output price (including thinking tokens)". The thinking guide says it plainly: "When thinking is turned on, response pricing is the sum of output tokens and thinking tokens." Each response reports the count in total_thought_tokens.
Defaults matter here. Carbon and Argon start at the medium thinking level; Hydrogen starts at high, and Google added an xhigh level above it. On Gemini 4, Google also bills internal tool system instructions and function declarations when you use grounding.
Watch max_output_tokens. If the model hits that cap while it is still reasoning, Google says it "stops generating with status "incomplete" and returns truncated or empty output (while still billing for any thinking tokens generated)." You pay for an answer you never got. Google's advice is to lower thinking_level instead of setting a small output cap.
To estimate before you pay, Google's tokens guide gives these sizes:
- Text: "a token is equivalent to about 4 characters. 100 tokens is equal to about 60-80 English words."
- PDFs: "Each document page is equivalent to 258 tokens", billed at the image token rate.
The countTokens call measures input only. You can also count a prompt's tokens before you send it with the free token counter on this site.
Batch, Flex and Priority change what the same token costs
Each model has four tabs on Google's pricing page, and Argon and Hydrogen add a fifth: Contributor. The same token costs a different amount in each. Google's own descriptions of those four, read 14 November 2026:
| Tier | Price vs Standard | Latency | Status and catch |
|---|---|---|---|
| Standard | Full price | Seconds to minutes | The default |
| Batch | 50% off | Up to 24 hours; Google says "in majority of cases, it is much quicker" | generateContent API only; its own limits, such as 100 concurrent batch requests and a 2GB input file |
| Flex | 50% off | 1 to 15 minute target | Preview; best-effort, on "sheddable" capacity |
| Priority | "75-100% more than Standard"; 1.8x on the price tables | Seconds | Preview; traffic over dynamic limits is downgraded to Standard |
Google's tier table and its Batch, Flex and Priority docs, read 14 November 2026. Flex and Priority arrived on 1 April 2026.
On Carbon, that spread is $1 input per 1M on Batch or Flex, $2 on Standard and $3.60 on Priority until 30 November, then $2, $4 and $7.20 from 1 December. Argon and Hydrogen also show a Contributor tab at Batch-level prices, but Google says it "is offered as a private preview on an invite-only basis", and prompts sent through it are used to improve Google's products.
Context caching is automatic on Gemini 4 and rented by the hour on older models
As of 14 November 2026, implicit caching is on by default for every Gemini 2.5 and newer model. Google's caching guide: "We automatically pass on cost savings if your request hits caches. You don't need to configure anything to turn it on." A hit costs 5% of the input price on Gemini 4 ("Implicit caching discounts cached input by 95%") and 10% on the 3.x models. On 3.8 Flash that is $0.075 against $0.75 per 1M until 31 December, then $0.15 against $1.50. Implicit hits carry no storage fee.
The rules that decide whether you get a hit, from the same guide:
- A prompt needs at least 4,096 tokens of shared prefix on Carbon, Argon, Hydrogen, 3.8 Flash, 3.6 Flash and 3.1 Pro Preview. Shorter prompts never hit.
- In Gemini 4 agent loops, "Gemini caches the full previous request only if the new request adds at least 2,050 tokens."
- Explicit caching, where you create a cache object and keep it alive, is "not supported in the Interactions API, or for Gemini 4 models".
- On the 3.x models it works through generateContent and charges storage by the hour, from $0.50 per 1M tokens an hour on 3.8 Flash until 31 December ($1.00 from 1 January 2027) up to $4.50 on 3.1 Pro Preview.
Is an explicit cache worth renting? Take a 100,000-token product manual on 3.8 Flash at the intro price. Keeping it cached costs $0.05 an hour, and each request that reads it saves $0.0675 against sending it fresh. So it pays once the manual is read about once an hour (0.74 reads, to be exact). Ten cached hours a day for 30 days costs $15 in storage, or $30 from January, when storage and input both double and the break-even stays where it was. My arithmetic, on Google's list prices.
Grounding with Google Search: 5,000 free queries a month, then $14 per 1,000
On Gemini 3 and newer models, Google's pricing page, as of 14 November 2026, says: "5,000 free search requests per month (shared across all Gemini 3 and newer models), then $14 per 1,000 requests (subject to a daily cap of 1,500 requests per day)." Read the footnote twice: "A customer-submitted request to Gemini may result in one or more queries to Google Search. You will be charged for each individual search query performed." You pay per search query, not per prompt.
The 1,500-a-day cap is per project across all paid tiers and resets at midnight Pacific time; for more, Google points to Vertex AI's default of 1,000,000 search queries a day. Grounding with Google Maps works the same way, 5,000 free a month and then $14 per 1,000. The 2.5 models keep an older rule: $35 per 1,000 grounded prompts after 1,500 free a day, and $25 per 1,000 for Maps. Google began billing Gemini 3 grounding on 5 January 2026.
The other tools bill as tokens, as of November 2026:
- Code execution is "billed at the standard token rates", and "no charges accrue for session runtime".
- URL context: "Charged as input tokens per model pricing."
- Deep Research, Managed agents and the Antigravity Agent bill every token of inference at standard list rates, intermediate reasoning included. For Managed agents and Antigravity, sandbox compute "is not billed during the preview period."
What a month of Gemini costs: three real workloads
Three common jobs, priced at Google's Standard list prices as of 14 November 2026 unless I say Batch. The volumes are my assumptions, stated in each case, and every total is my arithmetic. Your token counts will differ. The method holds.
A support chatbot costs $292.50 a month on 3.8 Flash until January.
Assume 30,000 conversations a month, each averaging 8,000 input tokens (system prompt, help-centre snippets and the growing history) and 1,000 output tokens, thinking included. That is 240M input and 30M output tokens.
| Model and date | Input | Output | Month |
|---|---|---|---|
| 3.8 Flash, until 31 Dec 2026 | $180.00 | $112.50 | $292.50 |
| 3.8 Flash from 1 Jan 2027, or Hydrogen on any date | $360.00 | $225.00 | $585.00 |
| 3.5 Flash-Lite | $72.00 | $75.00 | $147.00 |
My arithmetic on Google's Standard prices per 1M tokens, read 14 November 2026: 240M input and 30M output a month.
- Grounding: if 6,000 conversations trigger one search each, 1,000 queries pass the free 5,000 and add $14. That is 200 a day, far under the cap, though one prompt can trigger several queries.
- Caching, as an illustration only: if 5,000 of each conversation's 8,000 input tokens hit the implicit cache on 3.8 Flash's intro price, input falls from $180 to $78.75 and the month to $191.25. Nobody can promise that hit rate, and a shared prefix under 4,096 tokens never qualifies.
- The tier trap: $292.50 is over Tier 1's $250 monthly cap. Reach Tier 2 first ($100 paid and three days after the first payment), or service pauses until the 1st of the next month.
A document summariser costs under $14 a month on 3.5 Flash-Lite.
Assume 2,000 PDFs a month at 40 pages each. At Google's 258 tokens a page plus a 500-token instruction, each document is 10,820 input tokens, so the month is 21.64M input. A summary plus its thinking comes to 1,500 output tokens, or 3M a month.
- 3.5 Flash-Lite: $6.49 input plus $7.50 output, $13.99 a month.
- 3.8 Flash: $27.48 until 31 December, then $54.96 from 1 January 2027, which is also Hydrogen's price.
- 3.1 Pro Preview: $79.28, since every document sits far under 200,000 tokens.
- Gemini 4 Carbon: $73.28 in November and $146.56 from 1 December.
Run it overnight on Batch and every figure halves, so 3.8 Flash at the intro price drops to $13.74. A summariser that can wait up to a day has no reason to pay the Standard rate.
A nightly batch job spends most of its money on output tokens.
Assume 20,000 records a night, such as product descriptions to tag or tickets to classify, at 800 input and 200 output tokens each, for 30 nights. That makes 480M input and 120M output tokens a month, on Batch:
- 3.5 Flash-Lite: $72 plus $150, $222 a month, against $444 on Standard.
- 3.1 Flash-Lite: $150, but Google may retire it from 7 May 2027.
- 3.8 Flash: $405 until 31 December, $810 from 1 January 2027.
On every model, the 120M output tokens cost more than the 480M input tokens. On 3.5 Flash-Lite one output token costs as much as about eight input tokens, so a tighter answer format saves more than a shorter prompt. One more trap: a single night is 16M input tokens, above Tier 1's 10,000,000 enqueued-token cap for 3.5 Flash-Lite batches. Split each night into two jobs or wait for Tier 2, where the cap is 500,000,000. If your job runs around the clock instead, here is what an always-on agent costs to run.
How Gemini API billing works: prepaid credits and a $250 first cap
Google's billing page, read 14 November 2026, sets the rules a new paid account starts under:
- Prepay is the default for new accounts since 23 March 2026. The minimum purchase is $5, and "The maximum amount of credits you can prepay for is $5,000."
- Credits expire. Google says prepaid credits are non-refundable and expire after 12 months, and they pay for the Gemini API only.
- At $0, everything stops: "all API keys in all projects linked to that billing account will stop working simultaneously", with HTTP 402, and projects do not fall back to the free tier on their own. Google also warns that "Prepay billing is subject to overages during the approximately 10 minute billing pipeline latency", so a running batch job can overshoot, and the negative balance comes off your next purchase.
- Monthly caps by tier: $250 on Tier 1, $2,000 on Tier 2, and $20,000 to $100,000 or more on Tier 3. At the cap, service pauses for every project on the billing account until the 1st of the next month.
- A spend-rate limit: Tier 1 can spend $10 in any rolling 10 minutes (Tiers 2 and 3, $200), and going over returns
429 RESOURCE_EXHAUSTED. - The $300 Google Cloud Welcome credit does not help new accounts: "If you opened your account after March 2, 2026, you can't use these credits to pay for Gemini API and AI Studio usage".
- Failed requests: "If your request fails with a 400 or 500 error, you won't be charged for the tokens used. However, the request will still count against your quota."
Tier upgrades are automatic. Tier 2 needs $100 paid and three days since the first successful payment; Tier 3 needs $1,000 and 30 days, counted on your total Google Cloud spend, the Gemini API included. Google AI Studio itself stays free: "AI Studio usage remains free of charge unless users link a paid API key for access to paid features." And the API pricing, billing and terms pages do not mention Google AI Plus, Pro or Ultra at all: the API bills through a Cloud Billing account, apart from any Gemini app subscription.
Prepay offers auto-reload and a monthly auto-charge limit. If you switch on the first, set the second the same day: an agent stuck in a loop keeps billing thinking tokens until something stops it, and Google's own project-level spend caps are still marked "Experimental".

Images, video, voice and embeddings, one line each
Google prices its media models in their own units. Standard paid prices, as of 14 November 2026:
- Nano Banana 2.1 images: $30 per 1M image tokens, which Google puts at about $0.0336 for a 1K image and $0.113 for a 4K one.
- Nano Banana Pro: $0.134 per 1K or 2K image, $0.24 per 4K.
- Gemini Omni 1.2 Flash video: "approximately $0.07 per second" at 720p, so about $0.70 for a 10-second clip.
- Gemini 3.8 Flash TTS: about $0.00225 per 10 seconds of speech until 31 December 2026, then about $0.0045.
- Gemini 3.8 Live: audio at $0.005 a minute in and $0.018 a minute out.
- Gemini Embedding 2: $0.20 per 1M text tokens; the older
gemini-embedding-001costs $0.15. - Lyria 3.5 music: $0.08 a song.
The image, video and music models are paid only, while TTS, Live and both embedding models are free on the free tier. Veo and Imagen are no longer sold on the Developer API. Google shut down the Veo 3.1 previews there on 22 October 2026 with Omni 1.2 Flash as the replacement, points Veo's GA models to the Agent Platform, and lists Imagen 4 as shut down.
Gemini Enterprise Agent Platform, the old Vertex AI page, charges the same Global price per token
Gemini Enterprise Agent Platform is the enterprise route behind the Developer API's Enterprise card, and Google's old Vertex AI pricing address now redirects to its pricing page. For the models I checked on 14 November 2026, Global prices match the Developer API table above. Four differences:
- Non-global endpoints cost 10% more: Carbon is $2.20 / $11.00 at the intro price and $4.40 / $22.00 from 1 December, Hydrogen $1.65 / $8.25.
- The Carbon and Argon intro price arrives as a rebate. Google's footnote: "Promotional pricing provided through 50% credits back on net spend on select models within a given period. Promotional pricing does not apply to Gemini 4 Argon Cyber Permissive and Gemini 4 Carbon Cyber Permissive."
- From 13 November 2026, "When a completion exceeds 32K tokens in length, users will be charged for retrievable token usage even if the request ultimately fails due to client or server timeouts." That line is on the Agent Platform page only; the Developer API pages I read do not carry it.
- Grounding volume above the Developer API's 1,500 a day lives here, with the same 5,000 free queries a month and $14 per 1,000 after, plus grounding on your own data at $2.50 per 1,000 prompts.
The page also names Flexible Savings Plans, provisioned throughput and volume discounts. I read no figures for any of them, so the list price alone gives you no reason to move.
Gemini API vs OpenAI, Claude and DeepSeek on price
I lined up each vendor's top, middle and cheapest listed model by its own lineup order, from pricing pages read on 14 November 2026. Standard tier, short prompts, US dollars per 1M tokens:
| Position in lineup | Google Gemini API | OpenAI API | Anthropic Claude API | DeepSeek API |
|---|---|---|---|---|
| Top listed model | Gemini 4 Carbon: $2 / $10 to 30 Nov 2026, then $4 / $20 | gpt-6-astra: $10 / $50 | Claude Fable 5.2: $8 / $40 | deepseek-v4-pro: $1.32 / $3.96 peak, $0.66 / $1.98 off-peak |
| Middle | Gemini 4 Hydrogen: $1.50 / $7.50; 3.8 Flash: $0.75 / $3.75 to 31 Dec, then $1.50 / $7.50 | gpt-6.1-sol: $2 / $10; gpt-6.1-terra: $1 / $5 | Claude Opus 5.5: $4 / $20; Claude Sonnet 5.5: $2 / $10 | Two models only |
| Cheapest | 3.5 Flash-Lite: $0.30 / $2.50; 3.1 Flash-Lite: $0.25 / $1.50 (retires from May 2027 at the earliest) | gpt-6-luna: $0.10 / $0.50 | Claude Haiku 5.5: $0.10 / $0.50 for prompts up to 100K tokens | deepseek-flash: $0.30 / $1.20 peak, $0.15 / $0.60 off-peak |
| Batch | 50% off | 50% off | 50% off | Off-peak hours are half price instead |
Google's, OpenAI's, Anthropic's and DeepSeek's own API pricing pages, read 14 November 2026. Position follows each vendor's lineup order; it is no claim that the models are equivalent.
Now one basket: 100M input and 10M output tokens in a month. At those Standard prices, by my arithmetic, it costs:
- gpt-6-astra $1,500; Claude Fable 5.2 $1,200.
- Claude Opus 5.5 $600, the same as Gemini 4 Carbon from 1 December. Carbon costs $300 until 30 November.
- Gemini 4 Hydrogen $225.
- deepseek-v4-pro $171.60 at peak or $85.80 off-peak.
- Gemini 3.5 Flash-Lite $55, and deepseek-flash $42 at peak or $21 off-peak.
- gpt-6-luna and Claude Haiku 5.5, $15 each.
So Gemini's top model is cheaper per token than OpenAI's and Anthropic's top listed models, even after 1 December, and its cheapest current models cost more per token than gpt-6-luna and Claude Haiku 5.5. Two caveats, one sentence each. Tokenizers differ, so the same text is a different token count at each vendor. And a model's place in its vendor's lineup says nothing about quality: this ranks prices only. For the other two big labs, my Claude pricing guide covers Anthropic's plans and API, and my ChatGPT pricing guide does the same for OpenAI.
Each rival has its own surcharge to watch.
- OpenAI's "Long" context view doubles input: gpt-6-astra goes from $10 / $50 to $20 / $75. I could not read the token threshold for Long on the page.
- Claude Haiku 5.5 costs five times as much above 100K-token prompts, $0.50 / $2.50, and Anthropic charges 1.1x for US-only inference.
- DeepSeek's peak rates, double the off-peak ones, apply from 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays, excluding Chinese public holidays.
Which Gemini model should you pay for?
My rule, in order, at Google's list prices as of November 2026:
- Prototype on the free tier with nothing sensitive in it. Google may use it to improve its products, and human reviewers may read it.
- Ship high-volume, simple work on 3.5 Flash-Lite at $0.30 / $2.50 per 1M tokens, and put it on Batch wherever a day's delay is fine.
- Take 3.8 Flash for harder agent work while it costs $0.75 / $3.75, and budget for $1.50 / $7.50 from 1 January 2027. By then Hydrogen costs the same, and it is the newer model Google steers new projects to.
- Keep Carbon or Argon for the jobs that fail on cheaper models, and price them at $4 / $20 even if you start before the $2 / $10 intro ends on 30 November.
- Build nothing new on 2.5 models or 3.1 Flash-Lite. Move anything on 2.5 Pro before 1 December, or the routing moves it onto 3.8 Flash for you, breaking changes and all.
To price your workload across the labs at today's list prices, use the calculator. If you are paying for the app rather than the API, Google's consumer Gemini plans are in my other guide.
A note on how Gravity prices its own plans. Gravity, which publishes this blog, is not an API and does not sell tokens. It is for people who want a task done by a ready-made agent instead of building on a model API: you type the task, and an agent someone built runs it, once or on a schedule. Gravity is in public beta. Its plans are Free at $0 with no card, Mini Pilot at $15 a month, Auto Pilot at $20 a month and Auto Pilot Max at $100 a month, and you can buy more usage if you run out. Your first 3 agent runs are free.
FAQ
Is the Gemini API free?
Yes, on some models. As of 14 November 2026, Gemini 4 Hydrogen, 3.8 Flash, 3.6 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite and 3 Flash Preview are free on the free tier, within limits Google shows only in AI Studio, while Carbon, Argon, 3.1 Pro Preview and the image and video models are paid only. Google's terms say free-tier prompts may be used to improve its products and read by human reviewers.
How much does the Gemini API cost per 1M tokens?
As of November 2026, prices run from $0.25 input and $1.50 output per 1M tokens on 3.1 Flash-Lite, and $0.30 and $2.50 on 3.5 Flash-Lite, up to $2 and $10 on Gemini 4 Carbon and Argon until 30 November 2026, then $4 and $20 from 1 December. Gemini 3.8 Flash costs $0.75 and $3.75 until 31 December 2026, then $1.50 and $7.50, the same as Gemini 4 Hydrogen.
How much is Gemini 3 Pro?
Google shut down Gemini 3 Pro Preview on 9 March 2026, and the gemini-3-pro-preview name now points to Gemini 3.1 Pro Preview. As of November 2026, that model costs $2 per 1M input tokens and $12 per 1M output tokens for prompts up to 200,000 tokens, and $4 and $18 above that.
How much is Gemini Flash?
Gemini 3.8 Flash and 3.6 Flash cost $0.75 per 1M input tokens and $3.75 per 1M output tokens until 31 December 2026, then $1.50 and $7.50 from 1 January 2027. As of November 2026, Gemini 3.5 Flash-Lite costs $0.30 and $2.50, and 3.1 Flash-Lite costs $0.25 and $1.50.
Is Gemini 3.6 Flash free?
On the free tier, yes: Google lists its input, output and caching as free of charge, within limits shown in AI Studio. On the paid tier it costs the same as 3.8 Flash, $0.75 and $3.75 per 1M tokens until 31 December 2026, then $1.50 and $7.50.
Are thinking tokens billed?
Yes. Google bills thinking tokens at the output price, and every output price on its table includes them. If a response hits max_output_tokens while the model is still reasoning, you can get truncated or empty output and still pay for the thinking.
Will Gemini API prices go up?
Yes, and the dates are set. Gemini 4 Carbon and Argon go from $2 and $10 to $4 and $20 per 1M tokens on 1 December 2026, and 3.8 Flash and 3.6 Flash go from $0.75 and $3.75 to $1.50 and $7.50 on 1 January 2027. Google's terms let it change prices 30 days after posting them.
How does Gemini API billing work?
New accounts prepay from $5 into a Cloud Billing account, and Google says prepaid credits are non-refundable and expire after 12 months. At a $0 balance every API key on the account stops at once. Monthly spend caps rise by tier, from $250 to $2,000 to $20,000 or more.
Does a Google AI Pro subscription include the API?
Google's API pricing, billing and terms pages do not mention Google AI Plus, Pro or Ultra at all, and API use bills separately through a Cloud Billing account. Budget for the subscription and the API as two separate bills.
What does Batch save?
Batch is 50% off the Standard input and output price, with a 24-hour target turnaround, and it works with the generateContent API only. Flex gives the same 50% discount with a 1 to 15 minute target, but it is a preview with best-effort capacity.
How much does grounding with Google Search cost?
On Gemini 3 and newer models, 5,000 search queries a month are free, shared across all those models, and then Google charges $14 per 1,000 queries, as of November 2026. One prompt can trigger several queries, each one billed, and the Developer API caps grounding at 1,500 queries a day per project.
Is the Gemini API cheaper than OpenAI's?
At the top of the lineups, yes per token: Gemini 4 Carbon costs $2 and $10 per 1M tokens until 30 November 2026 and $4 and $20 from 1 December, against $10 and $50 for gpt-6-astra. At the bottom, no: gpt-6-luna costs $0.10 and $0.50, against $0.30 and $2.50 for Gemini 3.5 Flash-Lite. Tokenizers differ, so the same text is a different token count on each.
Sources
- Google, Gemini Developer API pricing (requested as ai.google.dev/gemini-api/docs/pricing), every model's Standard, Batch, Flex, Priority and Contributor tabs, no date shown on the page, read 14 November 2026; also rendered from India, same US-dollar prices.
- Google, Gemini models, read 14 November 2026.
- Google, Gemini API release notes, entries from 5 December 2025 to 12 November 2026, read 14 November 2026.
- Google, Gemini deprecations, read 14 November 2026.
- Google, Rate limits, read 14 November 2026.
- Google, Billing, including the tier table and FAQ, read 14 November 2026.
- Google, Gemini API Additional Terms of Service, effective 23 March 2026, read 14 November 2026.
- Google, What's new in Gemini 4 Carbon, Gemini 4 Argon and Gemini 4 Hydrogen, read 14 November 2026.
- Google, Gemini API docs on caching, Batch API, Flex, Priority, thinking, tokens, long decode and document processing, read 14 November 2026.
- Google Cloud, Agent Platform pricing (requested as cloud.google.com/vertex-ai/generative-ai/pricing) and Gemini Enterprise Agent Platform (requested as cloud.google.com/vertex-ai), read 14 November 2026.
- OpenAI, API pricing (requested as platform.openai.com/docs/pricing), Short and Long context views, read 14 November 2026.
- Anthropic, Claude API pricing and claude.com/pricing, read 14 November 2026.
- DeepSeek, API pricing, read 14 November 2026.
- Gravity pricing (gravity.fast/pricing), November 2026. Free at $0 with no card, Mini Pilot $15 a month, Auto Pilot $20 a month, Auto Pilot Max $100 a month.
- The three monthly bills, the caching illustration and break-even, the Carbon 1 December step and the 100M-in, 10M-out basket are my arithmetic on the figures above, shown in the body.


