Gemini 3.7 Flash is free in Antigravity and AI Studio, with a limited-time discount on OpenRouter. Here's every access point and price.
What is Gemini 3.7 Flash and why does its pricing matter? #
Gemini 3.7 Flash is Google’s latest fast, low-cost model in the Gemini line, released about three weeks after Gemini 3.6 Flash. It’s positioned as a workhorse for coding and agentic tasks rather than a frontier reasoning model. What makes it worth tracking isn’t just its capability jump over the previous Flash version, it’s that Google and third-party providers are currently making it available for free or at a steep discount across multiple platforms. For anyone building agentic tools, coding assistants, or high-volume AI features, that combination of decent performance and near-zero cost is the actual story.
TL;DR #
Gemini 3.7 Flash is free to use inside Google Antigravity, Google’s agentic coding environment, with no separate charge for the model itself.** AI Studio and AI Studio Build also offer it for free**, letting you chat, test prompts, or generate and deploy small apps without paying.** The standard API pricing is an introductory rate through the end of December**: 75 cents per million input tokens and $3.75 per million output tokens, roughly half of what 3.6 Flash cost at launch.** OpenRouter is running an additional limited-time discount until August 27**, dropping the price to around 38 cents per million input tokens and $1.88 per million output tokens, about 75% off the standard rate.Benchmark gains over 3.6 Flash are large for a point release, including notable jumps on long-horizon software engineering and agentic automation tests.** Free tiers are meant for testing, not production**, since Google can adjust quotas at any time, so heavy reliance on the free access points carries some risk for shipped products.
One coffee. One working app. #
You bring the idea. Remy manages the project.
How much does Gemini 3.7 Flash cost through the API? #
Outside the free access points, Google is charging an introductory API rate that runs until the end of December. That rate is 75 cents per million input tokens and $3.75 per million output tokens. That’s already roughly half the launch price of the previous Flash model, which is a meaningful cut for anyone running high-volume workloads like coding agents, chat applications, or batch processing jobs where token counts add up fast.
On top of Google’s own discounted introductory pricing, OpenRouter is offering an additional exclusive discount through August 27. During that window, the effective price on OpenRouter comes to about 38 cents per million input tokens and $1.88 per million output tokens, which works out to roughly 75% off the standard rate. For teams routing traffic through OpenRouter to tools like Cline, Roo Code, or Open Code, this is a narrow window to lock in agentic workflows at a fraction of typical cost. After August 27, pricing presumably reverts to Google’s standard introductory rate (or whatever OpenRouter sets going forward), so this discount is explicitly time-limited.
Where can you use Gemini 3.7 Flash for free? #
There are two confirmed free access points right now.
Google Antigravity: This is Google’s agentic coding environment, and Gemini 3.7 Flash is available there at no cost. You log in, select the model, and start building. For agentic coding specifically, getting a model of this quality without paying per-token is a strong deal, especially since agentic workflows tend to burn through tokens quickly with multi-step tool calls and long context.
AI Studio and AI Studio Build: Gemini 3.7 Flash is also free inside Google AI Studio, where you can chat with it directly or test prompts before wiring them into an application. AI Studio Build extends this further, letting you generate and deploy small apps using the model without a billing setup.
The caveat with both is that free tiers are explicitly intended for testing and experimentation. Google can change quotas or availability at any time, so it’s not a foundation to build a production service on. For personal projects, prototyping, or evaluating whether the model fits your use case, though, it’s more than sufficient.
Is Gemini 3.7 Flash actually good, or just cheap? #
Cheap and free access only matters if the model can do useful work, and the benchmark movement from 3.6 Flash to 3.7 Flash is unusually large for a release that shipped just three weeks after its predecessor. On Deep Sweep, a benchmark for long-horizon software engineering tasks, the model scores 65.3% compared to 49% for 3.6 Flash, a 16-point jump. On Frontier Code, focused on production-quality code output, it improves from 34.4% to 43.6%. On Web Dev Arena, an Elo-style comparison benchmark, it moved from around 1538 to about 1588. The biggest jump shows up on Automation Bench, an agentic benchmark, where the score nearly doubled, going from around 17% to roughly 30%.
- ✕a coding agent
- ✕no-code
- ✕vibe coding
- ✕a faster Cursor
The one that tells the coding agents what to build.
Those numbers matter less on their own than what they suggest: a model that got meaningfully better at multi-step agentic tasks and real coding work in a very short iteration cycle. In practical use, that shows up as the model being able to plan a sequence of actions, call tools correctly, and follow through on longer chains of steps without losing track, which is exactly the kind of reliability that agentic coding tools depend on.
On raw performance, the model reportedly runs at around 250 tokens per second and supports a 1 million token context window, putting it in a category where speed and context length both support agentic and coding use cases that require processing large codebases or long conversation histories.
Why is Gemini 3.7 Flash flying under the radar? #
Despite the pricing and benchmark story, this release hasn’t generated much buzz. A few likely reasons stand out. Flash models generally get less attention than Pro or frontier-tier releases, since most of the industry conversation centers on whichever model is claiming the top spot on reasoning or general intelligence benchmarks. A three-week gap between 3.6 Flash and 3.7 Flash also creates an assumption that this is a minor, incremental update rather than a substantial capability jump, even though the benchmark deltas suggest otherwise.
There’s also a structural reason cheap, fast models get overlooked: the “wow” factor of a new frontier model launch is easy to market, while a mid-tier model getting significantly better at agentic tasks is a quieter, more operational kind of upgrade. But for people actually building products, that operational upgrade is often more consequential. When a fast, inexpensive model becomes capable enough to handle real coding and agentic work, it changes what’s economically viable to run at scale, since you’re no longer forced to route every task through an expensive frontier model just to get reliable results.
Is Gemini 3.7 Flash worth using right now? #
For coding and agentic workflows specifically, yes, largely because of the combination of free access points and the temporary API discount. If you’re experimenting or building personal projects, Antigravity and AI Studio give you a no-cost way to test the model’s coding and agentic capabilities directly. If you’re already running API-based workflows through tools like Cline, Roo Code, or similar agentic frameworks, routing through OpenRouter before the August 27 discount ends gets you meaningfully cheaper inference for a model that’s already priced below its predecessor. The caveat is durability. Free tiers can shrink or disappear, and the OpenRouter discount has a hard expiration date. Anyone building something long-term should treat the current pricing as a limited window to evaluate the model cheaply, not as a permanent cost structure to design a business around.
Frequently Asked Questions #
Is Gemini 3.7 Flash completely free to use?
It’s free inside Google Antigravity and Google AI Studio (including AI Studio Build), but API access outside those environments is priced, currently at an introductory rate of 75 cents per million input tokens and $3.75 per million output tokens through the end of December.
What is the OpenRouter discount on Gemini 3.7 Flash?
OpenRouter is offering an additional discount on top of Google’s introductory pricing, bringing costs down to roughly 38 cents per million input tokens and $1.88 per million output tokens. This discount is scheduled to end on August 27.
How does Gemini 3.7 Flash compare to Gemini 3.6 Flash?
Remy doesn't write the code. It manages the agents who do. #
Remy runs the project. The specialists do the work. You work with the PM, not the implementers.
It shows notable benchmark improvements, including a jump from 49% to 65.3% on the Deep Sweep software engineering benchmark and roughly doubling the agentic automation score from about 17% to 30%, despite releasing only three weeks after 3.6 Flash.
Can I use Gemini 3.7 Flash’s free tier in production?
It’s not recommended. Google’s free tiers in Antigravity and AI Studio are intended for testing and experimentation, and quotas can change without notice, making them unsuitable as a stable foundation for production applications.
What is Gemini 3.7 Flash best used for?
Based on its benchmark profile and the demonstrated use cases, it’s positioned for coding tasks, agentic workflows involving multi-step tool use, and general day-to-day tasks where near-frontier performance at low cost matters more than absolute top-tier reasoning ability.