Gemini 3.7 Flash arrives with benchmarks Google graded itself — and a price that doubles in January
Google shipped a new "workhorse" model three weeks after the last one. The coding numbers are all first-party, the launch price is introductory, and UK users cannot reach it in the Gemini app at all yet.
Google announced Gemini 3.7 Flash on 13 August, calling it its most intelligent “workhorse” model and pointing at large gains in coding, tool use and agentic work. It arrived three weeks after Gemini 3.6 Flash. Before repeating the numbers, it is worth being clear about who produced them and what the small print says.
The benchmarks are Google’s own
The headline case for 3.7 Flash is coding. Google reports the DeepSWE v1.1 software-engineering score rising from 49.0 to 65.3, a FrontierCode result of 43.6 against the predecessor’s 34.4, an AutomationBench jump from 17.0 to 30.4, and a WebDev Arena Elo up from 1538 to 1588. The model card lays them out cleanly.
These are real, specific and checkable figures — and they are all first-party. They are Google grading Google’s model, on Google’s chosen benchmarks, against Google’s previous model. That is normal for a launch, but it is not independent evidence. Benchmark gains, particularly on coding suites that can leak into training data, have a habit of shrinking once outside labs and everyday users run their own tasks. The number that matters is the one you get on your own workload, and none of the launch material can supply that.
The context stays the same as 3.6 Flash: a roughly one-million-token input window and a 64,000-token output ceiling. The pitch is not more context; it is more competence per token in the same envelope.
Read the pricing before the benchmarks
Gemini 3.7 Flash launches at $0.75 per million input tokens and $3.75 per million output tokens — around half what the previous Flash cost at its own launch, and an obviously aggressive number aimed at developers building agents that burn tokens at scale.
It is also, in Google’s own words, introductory. That rate holds until 31 December 2026. On 1 January 2027 it doubles, to $1.50 and $7.50. Anyone sizing a product around the current price is budgeting against a figure with a four-month shelf life. Cheap-to-adopt and cheap-to-run are being quietly presented as the same thing here, and they are not.
The bit British readers should note
Availability is uneven, and the UK is on the wrong side of it. Developers can reach the model now through the Gemini API and Google AI Studio, and it is rolling into enterprise tools including Android Studio and the Gemini Enterprise platform.
In the consumer Gemini app, it is arriving through a feature Google calls Spark, behind an AI Pro or Ultra subscription. But UK users cannot currently reach 3.7 Flash in the Gemini app at all, regardless of what they pay — only the developer API route is open here. So the honest summary for a British reader is: if you write code against the API, you can use it today; if you are a paying Gemini subscriber expecting the newest model in the app you actually opened, you are waiting, with no date attached.
That gap between “launched” and “launched where you are” is the sort of detail a press release is built to skate over. The model may well be a genuine step up. It is still worth insisting that the evidence for that is currently Google’s alone, the price is temporary, and the app most people use here does not have it yet.