Gemini API Pricing 2026: Free Tier, Per-Model Rates, and How Startups Get It Free
Every Gemini model Google bills for in 2026, what the free tier really includes, the batch and caching discounts that matter, and the Google for Startups credits that make the whole thing free for early-stage companies.
Gemini API pricing per model (September 2026)
Standard paid-tier rates per million tokens (MTok) from Google's Gemini API pricing page. "Free" in the free-tier column means free input and output tokens through Google AI Studio, subject to rate limits. Batch is 50% off across the board.
| Model | Free tier | Input / MTok | Output / MTok | Batch (in / out) | Best for |
|---|---|---|---|---|---|
| Gemini 3.8 Flash | Yes | $0.75 | $3.75 | $0.375 / $1.875 | Current default for most apps. Promo rate through Dec 31 2026, then $1.50 / $7.50. |
| Gemini 3.7 Flash | Yes | $0.75 | $3.75 | $0.375 / $1.875 | Previous Flash, same price |
| Gemini 3.6 Flash | Yes | $0.75 | $3.75 | $0.375 / $1.875 | Older Flash, same price |
| Gemini 3.5 Flash | Yes | $1.50 | $9.00 | $0.75 / $4.50 | No promo discount; pricier than the 3.6+ line |
| Gemini 3.5 Flash-Lite | Yes | $0.30 | $2.50 | $0.15 / $1.25 | High-volume classification and extraction |
| Gemini 3.1 Flash-Lite | Yes | $0.25 | $1.50 | $0.125 / $0.75 | Cheapest current-generation model |
| Gemini 3.1 Pro Preview | No | $2.00 (≤200k) / $4.00 (>200k) | $12.00 / $18.00 | $1 / $6 (≤200k) | Hardest reasoning tasks; long-context surcharge above 200k |
| Gemini 2.5 Pro | Yes | $1.25 (≤200k) / $2.50 (>200k) | $10.00 / $15.00 | $0.625 / $5.00 (≤200k) | Previous-generation Pro, still free-tier eligible |
| Gemini 2.5 Flash | Yes | $0.30 | $2.50 | $0.15 / $1.25 | Cheap, fast, widely deployed |
| Gemini 2.5 Flash-Lite | Yes | $0.10 | $0.40 | $0.05 / $0.20 | The cheapest text model Google sells |
| Gemini Omni Flash | No | $1.50 | $9.00 text / $17.50 video | — | Multimodal generation including video output |
A few pricing details that change the bill:
- Context caching costs between $0.025 and $0.40 per MTok depending on model, plus a storage charge (for example $0.50 per hour on 3.8 Flash). Worth it for any prompt with a large fixed prefix.
- Long context surcharge. Pro models charge more per token for requests above 200k tokens. Flash models do not.
- Grounding with Google Search includes 5,000 free search requests a month on Gemini 3.x models, then $14 per 1,000.
- Priority tier is typically 1.8x the standard rate for guaranteed capacity.
What the free tier actually includes
Google's free tier is the most generous of the major model providers, and it is the reason a lot of prototypes get built on Gemini first. The important points:
- Free input and output tokens on every Flash and Flash-Lite model in the table, plus Gemini 2.5 Pro. You do not pay per token; you are rate-limited instead.
- Rate limits are per model and measured in requests per minute and per day. They are tuned for development and low-volume apps. Exceed them and requests fail rather than being billed.
- Your data may be used to improve Google's products on the free tier. That is the trade. Paid tier requests are not used for training. For anything touching customer data, move to paid before launch.
- No card required to start. Create a key in Google AI Studio and you are calling the API in a minute.
The free tier is not the same thing as Google Cloud credits. The free tier is Google AI Studio's developer allowance. Credits are money on a Google Cloud billing account that pays for Vertex AI usage at the paid-tier rates above, with paid-tier data handling. Startups want the second one.
Gemini vs Claude vs OpenAI on price
Per million tokens, Gemini is the cheapest of the three at every tier. Gemini 3.8 Flash at $0.75 / $3.75 sits under Claude Haiku 4.5 ($1 / $5) and well under Claude Sonnet 5 ($2 / $10). Gemini 3.1 Pro at $2 / $12 is priced between Sonnet 5 and Opus 5 ($5 / $25), and cheaper than OpenAI's gpt-5.5 ($5 / $30).
Price per token is not price per task, though. Three things to check on your own workload:
- Tokens per task. Different tokenizers count the same text differently. Compare total cost for a fixed batch of real prompts, not list price.
- Retries and quality. A cheaper model that needs a second pass is not cheaper. Measure the whole pipeline.
- Which credits you hold. If you have Google Cloud startup credits, Gemini on Vertex AI is free until they run out. If you have Claude credits, Claude is. Route by what is funded first.
For founders
How startups get Gemini free
Google does not run a separate Gemini credit program for startups. Instead, the Google for Startups Cloud Program gives early-stage companies Google Cloud credits, and Gemini on Vertex AI is billed as Google Cloud usage. Credits cover it at the paid-tier rates in the table above, with paid-tier data handling.
The fine print
The standard track is up to $200,000 in credits over two years for startups that have raised outside funding. The AI track goes up to $350,000 for AI-first companies and adds things like Gemini and Vertex AI training and partner introductions. Self-funded and pre-seed founders get a smaller starting allocation and can scale up as they raise.
What the credits do not cover: Google AI Studio's free tier is separate and unaffected, and the credits will not pay for Gemini via a third-party gateway that bills you directly. Move production traffic to Vertex AI with the same models before you turn the credits on.
Google for Startups Cloud Program
Up to $350,000
- Up to $200,000 in Google Cloud credits for 2 years
- Access to Google Cloud experts and technical guidance
- Google Cloud training credits
Step by step: applying for Google Cloud credits and putting them on Gemini
The application is straightforward. The part people get wrong is the Vertex AI setup afterwards, so the last two steps matter as much as the first three.
-
1
Check which track you qualify for
Pre-seed and self-funded startups qualify for the starting tier. Startups with equity funding qualify for the up-to-$200,000 track. AI-native companies with funding can apply for the AI track. Our Google Cloud program page lists the current eligibility and common rejection reasons. -
2
Apply at cloud.google.com/startup
You need a Google Cloud billing account, a company website, and details of your funding. Referral from a Google partner VC or accelerator speeds review and unlocks the higher tiers. Review typically takes days to a few weeks. -
3
Move Gemini calls from AI Studio to Vertex AI
Same models, different endpoint and auth. Vertex AI bills to your Google Cloud project, which is where the credits sit. AI Studio keys bill separately and will not draw on credits. -
4
Turn on context caching and batch
Cache large fixed prompts and route anything asynchronous through the Batch API for 50% off. Credits last twice as long. -
5
Set a budget alert before the credits run out
Google Cloud budgets can email you at 50%, 90% and 100% of credit spend. When they are gone, billing switches to your card at the same rates with no migration, so you want warning.
Free tier vs paid vs startup credits: the math
For a seed-stage startup running a Gemini-powered product at roughly 500 million tokens a month (an even split of input and output on 3.8 Flash), here is what a year costs on each route.
| Route | Year-one cost | What it covers | Where it breaks |
|---|---|---|---|
| AI Studio free tier | $0 | Prototyping, demos, low-volume internal tools | Rate limits stop you well before production volume. Data may be used for training. |
| Paid tier, Gemini 3.8 Flash | About $13,500 | Production usage at list price | Rises to about $27,000 when the promo rate ends after 2026. |
| Paid tier with batch and caching | About $7,000–$9,000 | Same usage, async work batched, prompts cached | Only the async share gets the batch discount. |
| Google for Startups credits on Vertex AI | $0 until credits are used | Production Gemini usage at paid-tier rates and data handling | Credits have a two-year window and only cover Google Cloud billing. |
Most funded startups should run the third and fourth rows together: Vertex AI on credits, with batch and caching turned on so the credits cover more months.
Questions founders ask
Is the Gemini API free?
How much does Gemini 2.5 Flash cost?
How much does Gemini Pro cost per million tokens?
Is Gemini cheaper than Claude or OpenAI?
Can startups get Gemini API credits?
What is the difference between Google AI Studio and Vertex AI pricing?
Does the Gemini Batch API really cost half?
Related guides
Google Cloud startup credits
Eligibility, tracks, and how to apply.
Claude API pricing
Anthropic's per-token rates for comparison.
OpenAI API pricing
GPT model rates and OpenAI startup credits.
Claude pricing
Every Claude plan and the startup program.
Cloud credits for students
What students can get from Google Cloud and others.
All AI startup credits
Every AI credit program in the directory.
