Google pushed Gemini 3.7 Flash to general availability on August 13, three weeks after the model it replaces. DeepSWE v1.1 climbs from 49.0% to 65.3%, and the launch price sits at $0.75 per million input tokens until December 31.
Key Takeaways
- Gemini 3.7 Flash hit general availability on August 13, three weeks after 3.6 Flash
- Launch pricing runs at $0.75 / $3.75 per million tokens through December 31, 2026, then doubles on January 1
- The sharpest gains land on code and automation, with AutomationBench moving from 17.0% to 30.4%
Have an AI Sum Up This Article
ChatGPTTwenty-one days between two Flash releases
Google did not let its workhorse model age. Gemini 3.7 Flash lands in general availability three weeks after 3.6, with the same positioning on paper and different numbers almost everywhere else.
The figures published in the model’s official launch page spell out the delta. FrontierCode 1.1 Main rises from 34.4% to 43.6%. DeepSWE v1.1 goes from 49.0% to 65.3%. WebDev Arena Elo picks up fifty points, 1538 to 1588.
Two results outside coding deserve a closer read. On GDP.pdf, a dense document comprehension test, the model moves from 22.0% to 34.0%. On AutomationBench, it goes from 17.0% to 30.4%.
The rest of the spec sheet holds still. One million tokens of context, 64,000 output tokens at most, tunable thinking levels across low, medium and high, and input across text, image, speech and video. The same built-in toolset as 3.6 Flash, with nothing added.
That combination is worth naming for what it is. Google kept the envelope untouched and spent the three weeks on capability inside it, which is the cheapest kind of upgrade to adopt because nothing downstream has to be rewired. Context budgets, tool wiring and output limits all carry over unchanged.
The contrast with the Pro line is hard to miss. While Flash ships twice in three weeks, Gemini 3.5 Pro slipped again while Google rebuilt it. The workhorse moves, the flagship waits, and the tempo gap between the two lines is turning into a fixture of the house.
What the launch price does to a product team’s math
Pricing is the real story here. Gemini 3.7 Flash bills at $0.75 per million input tokens and $3.75 per million output tokens, and that rate expires on December 31, 2026. From January 1, 2027, it becomes $1.50 and $7.50.
A team migrating today is therefore building its cost model on a number that doubles in four and a half months. The question is less whether the switch pays off now, and more whether it still holds at the January rate.
That answer depends on the workload shape. On coding tasks where first-pass accuracy improves, better output absorbs the doubling because retries drop. On high-volume document processing, the increase shows up straight on the invoice.
Introductory pricing as a weapon is familiar ground at Google. The consumer tier already played that card when Google AI Plus dropped to $4.99 to open a price war. The API version runs the same play, with an expiry date printed on it.
On distribution, the model reaches developers through Google AI Studio, Android Studio and Google Antigravity, enterprises through the Gemini Enterprise Agent Platform, and consumers through Gemini Spark for Google AI Pro and Ultra subscribers across more than a hundred and sixty countries. CBRN and cyber offense safeguards were refreshed alongside the release.
More articles on Horizon
- Kimi K3 Test Ranks It First on Frontend Code
- Twitch Trains Amazon AI on Your Streams by Default
- Grok 4.6 Matches the Best Models at Half the Price
January’s calendar puts rivals on a clock
A release every twenty-one days on the most heavily used tier of the market is a cadence few labs can sustain. The signal sent to competitors is about frequency more than benchmarks.
Google is also pushing on parallel tracks. Its own research keeps testing alternative architectures, as it did when DiffusionGemma wrote text four times faster. The catalogue widens while the main line accelerates.
Distribution remains the heaviest lever. A model shipped on day one to the Pro and Ultra subscribers of a service that closed in on one billion monthly users does not need to win a benchmark to become the default choice.
The automation score is the number rivals should read first. Doubling AutomationBench in three weeks says Google is optimising this tier for agent workloads rather than chat, and the cheap tier is exactly where agent traffic concentrates because those workflows fire thousands of calls per task.
For rivals, January is the window worth watching. If Google doubles its rates as stated, the price gap against Chinese open-weight models and mid-tier American lines narrows on its own. That is where the loyalty of teams migrating today gets tested.
One question the spec sheet leaves open. A cadence this fast forces teams to re-evaluate their model every three weeks, with all the validation work that carries. At some point, release speed becomes a cost for whoever absorbs it as much as an edge for whoever sets it.
Follow the story on Horizon.
