Google Gemini API Review 2026

What is Google Gemini API?

Gemini 3.8 Flash, Google's newest Flash model, costs nothing to call on the free tier, while the flagship Gemini 4 Argon, generally available since 13 October 2026, works only with billing switched on. Below are the gemini api free tier limits and pricing as Google's official pricing page lists them in October 2026, followed by keys, quotas, data rules and the nearest rivals, so a founder or agency in India can estimate a monthly bill before writing much code.

All prices are in US dollars for the Standard tier. Google publishes no rupee rates for the API, so Indian teams should budget in dollars with a buffer for currency swings.

Which Gemini models cost nothing to call

Besides 3.8 Flash, the free tier covers 3.6 Flash, both Flash-Lite generations, the 3.8 Live voice models, the 3.5 Transcribe speech-to-text models and the 3.8 text-to-speech voices. The older 2.5 family is also free of charge; see the access catch below. What you cannot get without billing: Argon, 3.1 Pro Preview, the Nano Banana image models and the Omni video model.

Context caching is listed as unavailable on free tier rows. Grounding with Google Search is also off for the free 3.x Flash models (the Live models are the exception), which limits free prototypes that need fresh web results. Typing prompts inside Google AI Studio in the browser is free as long as you have not linked a paid key to that project.

Gemini API free tier limits and pricing tiers, quota by quota

Every model has three main ceilings: requests per minute (RPM), input tokens per minute (TPM) and requests per day (RPD). Breaking any one of them returns an error. Google does not print per-model free numbers in the docs; each project's live limits sit on the rate limit screen in AI Studio, so check there.

The daily counter resets at midnight Pacific time. In India that is 12:30 pm IST while the US is on daylight saving and 1:30 pm IST in winter.

Tier 1 needs an active billing account and carries a $250 cap per monthly billing cycle. Tier 2 requires $100 already paid plus three days since the first payment, with a $2,000 cap. Tier 3 asks for $1,000 and 30 days, and its cap runs from $20,000 to over $100,000. There is also a spend-based limit over a rolling 10-minute window: $10 on Tier 1, $50 on Tier 2 and $200 on Tier 3.

Clearing a 429 RESOURCE_EXHAUSTED error

The "quota exceeded" message almost always means one of four things. You crossed RPM, TPM or RPD on that model. You hit the 10-minute spend ceiling. Your prepaid balance reached zero, which shows up as HTTP 402 instead. Or your billing account touched its tier cap, in which case every linked project pauses until the cycle restarts on the 1st (Google has a form for requesting a higher cap).

Google's troubleshooting page recommends exponential backoff with random jitter and a maximum retry count, and says to retry only on 429, 408 and 5xx codes. The official Python SDK already retries these up to four times. Other fixes: spread calls across the minute, send bulk work through the Batch API, drop to a Flash-Lite model for simple jobs, or attach billing to move off the free tier.

Your Gemini API key: creating one in Google AI Studio

Open the API keys page at aistudio.google.com/apikey and accept the terms. New users get a default Cloud project and a key automatically. If you already run Google Cloud, import an existing project from Dashboard and then Projects, and create the key inside it.

• The SDKs read GEMINI_API_KEY or GOOGLE_API_KEY from the environment; if both exist, GOOGLE_API_KEY wins.
• Since 28 May 2026, new keys made in AI Studio are authorization keys tied to a service account. Older standard keys keep working only if they carry explicit restrictions; the API rejects unrestricted ones.
• Restrict each key to the Gemini API only. Long-dormant unrestricted keys have been blocked since 7 May 2026.
• Never bundle the key inside an Android or web app. Route calls through your own backend.

To pay, click Set up billing on the project, choose your country and add a card. New accounts default to prepay: a minimum top-up of $5, a maximum of $5,000, and credits that expire after 12 months. The $300 Google Cloud welcome credit cannot be spent on Gemini API usage for accounts opened after 2 March 2026.

Gemini 2.5 Flash vs Pro pricing, and why new projects skip both

The 2.5 generation is still on the price list. Gemini 2.5 Pro charges $1.25 per million input tokens and $10 per million output tokens for prompts up to 200k tokens, rising to $2.50 and $15 for longer prompts. Gemini 2.5 Flash charges $0.30 input ($1.00 for audio) and $2.50 output, and 2.5 Flash-Lite is the cheapest text option Google lists at $0.10 and $0.40. Pro therefore costs roughly four times Flash on both input and output.

The problem is access. Since 18 September 2026, the API changelog says 2.5 models are available only to users who have called them recently. They are not deprecated and have no shutdown date, but a brand-new project should follow Google's advice and pick 3.5 Flash-Lite or 3.8 Flash.

Caching, batch jobs and grounding: the levers on your bill

Three switches change cost more than model choice. Context caching stores a large repeated prompt so later calls pay a fraction; on 3.8 Flash, cached input costs $0.075 per million tokens through 2026 and $0.15 from 1 January 2027, plus an hourly storage fee, while Argon supports only implicit caching. The Batch API halves the rate for jobs that can wait, so Argon drops to $2 input and $10 output. Grounding with Google Search gives 5,000 free requests a month shared across Gemini 3 and newer models, then $14 per 1,000. Google bills each search query the model runs, so one grounded prompt can count more than once.

Function calling, structured JSON output, code execution, URL context and file search all work on Argon. Multimodal input covers text, images, video, audio and PDFs on both Argon and 3.8 Flash, each with a long context window of 1,048,576 tokens. Argon returns up to 65,536 output tokens per synchronous call, or up to a million with long decode. The Live API runs on 3.8 Live, with audio input around $0.005 a minute. Embeddings come from gemini-embedding-001 and the multimodal Embedding 2 preview.

Gemini API vs OpenAI API on price per million tokens

OpenAI's pricing page shows gpt-6.1-sol at $2 input and $10 output, half of Argon's regular $4 and $20. OpenAI's top gpt-6-astra costs $10 and $50, and its small gpt-6-luna undercuts every current Gemini Flash-Lite at $0.10 and $0.50. The bigger gap is free access: OpenAI does have a free usage tier, but none of those three GPT-6 models runs on it, whereas Google's free tier includes its newest Flash.

Google runs an OpenAI-compatible endpoint, so existing OpenAI SDK code needs only a new key, a new base URL and a Gemini model name. Gemini wins on the free tier, native video and audio understanding and Search grounding; OpenAI wins on a cheaper mid-tier model and a cheaper bottom tier.

AI Studio key or Vertex AI: picking the route for a company account

The Gemini Developer API (keys from AI Studio) and the Google Cloud route, long known as Vertex AI and now labelled the Gemini platform in Google's docs, serve the same models through the same Google GenAI SDK; only the client setup differs. The Cloud route authenticates with service accounts and brings enterprise controls, and its regional availability can differ from the Developer API. Google advises most developers to start on the Developer API and move only for those controls. Tuned models built in AI Studio must be retrained if you move.

What Google does with free tier prompts

Under the Gemini API terms, content sent to unpaid services can be used to improve Google's products and models, and human reviewers may read it after it is disconnected from your account. Google tells users not to send sensitive, confidential or personal information on the free tier. Paid usage is not used for training; prompts are logged for a limited period for abuse detection and handled under Google's Data Processing Addendum. Search and Maps grounding keep prompts for 30 days.

Other rules: you must be 18 or older, apps for users in the EEA, Switzerland or the UK must use paid services, and Google claims no ownership of outputs. The terms frame the API for professional or business use and bar using it to build competing models. India is on the list of supported countries.

Free alternatives if the Gemini API quota runs out

• OpenRouter free models: IDs ending in :free allow 20 requests a minute and 50 a day, or 1,000 a day once you have bought 10 credits.
• Groq: its rate limit table lists free plan limits such as 30 requests a minute and 1,000 a day for gpt-oss-120b.
• Gemma: Google's open-weight models can run on your own GPU at zero API cost, which suits offline or privacy-sensitive work.
• DeepSeek: no free API credits, but its Flash model starts at $0.15 input in off-peak hours.

A sensible first fortnight with the API

Prototype in AI Studio on 3.5 Flash-Lite or 3.8 Flash using test data only. Log token counts per request from day one, because output tokens (including thinking) drive most bills. Set a project spend cap in AI Studio before anyone shares the app. Turn on billing before real user data arrives, mainly for the data terms rather than the speed. Move nightly summarisation or tagging jobs to batch mode. Cost any 3.8 Flash plan at its 2027 rate of $1.50 / $7.50.

Teams that need a fixed monthly price, or that cannot absorb a repricing of the model they ship on, should keep an OpenAI or open-weight fallback ready.