Gemini 3.7 Flash is the workhorse model of Google's Gemini 3 family, built for coding, agents and document-heavy work. It shipped a mere three weeks after Gemini 3.6 Flash, with API rates cut in half for a limited launch window. Developers can run it right away in AI Studio, Antigravity and the Gemini API, while Google AI Pro and Ultra subscribers get it inside Spark and AI Mode in Search.
Gemini 3.7 Flash posts a sharp jump over Gemini 3.6 Flash on software engineering tests, hitting 65.3% on DeepSWE v1.1 versus 49% three weeks earlier. Google also reports more complete web apps in fewer prompts and steadier reading of dense finance, legal and bioscience documents.
In practice, you paste a hundred-page PDF or a full code repository into the one-million-token window, and the model plans, fixes and double-checks without losing the thread. Three thinking levels can be tuned per task (medium is the default, and it covers most situations).
| Benchmark | Gemini 3.6 Flash | Gemini 3.7 Flash |
|---|---|---|
| DeepSWE v1.1 | 49.0% | 65.3% |
| FrontierCode 1.1 Main | 34.4% | 43.6% |
| WebDev Arena (Elo) | 1538 | 1588 |
| AutomationBench | 17.0% | 30.4% |
| GDP.pdf | 22.0% | 34.0% |
Google slots this model well below its direct rivals, with Claude Sonnet 5 and GPT-5.6 Terra listed at two to three times the per-million-token rate in the launch comparison chart. The in-house numbers stay honest, though, since Terra keeps the lead on Terminal-bench and Sonnet 5 wins some agent tasks.
One detail got people talking, the release landed before Gemini 3.5 Pro, which still has no date. The Flash line is now where Google iterates fastest, and the LLM model ranking is a handy way to follow a pecking order that shifts month to month.
Free hands-on testing starts in Google AI Studio, straight from the browser. For production or agentic work, several doors open from there.
On the API side, launch pricing sits at $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026, before moving to $1.50 and $7.50. Numbers like these rarely sit still, so treat the official pricing page as the final word.
Yes for testing, through Google AI Studio, no credit card required. API usage is then billed per token, and inside the Gemini app the model powers the Spark agent, which sits behind Google AI Pro and Ultra plans, as does AI Mode in Search.
Same specs, reworked engine. The one-million-token window, the 64k output cap and the built-in tools carry over unchanged, but scores climb steeply in debugging, web app building and business workflow automation. The introductory API rate applies to both versions until the end of 2026.
Not across the board. Google's own published numbers show it ahead of GPT-5.6 Terra on several coding tests, while Terra holds Terminal-bench and OSWorld, and Claude Sonnet 5 stays ahead on some agent evaluations. Its strongest argument is the ratio between those scores and its price.
No date has been given. Google shipped 3.7 Flash before its next flagship, and press reports noted the company declined to provide a timeline for Gemini 3.5 Pro even though partner testing had been mentioned earlier. For now, the Flash line is where fresh engineering lands first.
Verdict: Sixteen points gained on DeepSWE and an API bill halved through the launch window, that math speaks for itself for teams running high-volume coding and business agents. Solo developers get a rare shot at near-flagship results for Flash-tier money.
Claude Fable 5 
Grok 4.6 
Claude Opus 4.8 
