1300 166 138 Tell us what you need
News / Google

Google launches Gemini 3.7 Flash at half price for agents

By Monika Tantau • Published 3 months ago
Google launches Gemini 3.7 Flash at half price for agents

Key takeaways

• Google released Gemini 3.7 Flash on 13 August with introductory pricing until 31 December
• Input costs about $1.05 per million tokens now and roughly doubles from 1 January 2027
• FrontierCode coding score jumps from 34.4 to 43.6 per cent over Gemini 3.6 Flash
• Spark reaches more than 160 countries including Australia, while the EEA misses out for now

Most Read

Google has released Gemini 3.7 Flash, billing it as its most intelligent workhorse model yet for coding and agents. The launch landed on 13 August, only three weeks after Gemini 3.6 Flash, and it comes with introductory pricing set at roughly half the standard rate until 31 December, according to Google.

The workhorse tier is where the real fight is now. Agent workloads burn through millions of tokens on ordinary business tasks, so the cost and speed of a vendor’s mid tier model matters more to most companies than whatever sits at the top of the range. OpenAI, Google and Anthropic all ship a cheap fast tier next to their flagships these days, and those tiers are where most production automation actually runs. Google has priced this release to be tried, and it has published the date the discount ends.

What Google shipped and what it costs

The benchmark jumps over Gemini 3.6 Flash are large for a three week gap. FrontierCode 1.1, a software engineering test, moves from 34.4 to 43.6 per cent. DeepSWE climbs from 49.0 to 65.3 per cent, WebDev Arena rises 50 Elo points to 1588, and AutomationBench, which measures agent style work, nearly doubles from 17.0 to 30.4 per cent. A knowledge work test run over PDF documents also climbs from 22.0 to 34.0 per cent, which is the score most relevant to document processing.

Introductory pricing runs until 31 December at about $1.05 per million input tokens and $5.30 per million output tokens, which Google says is half what the original 3.6 Flash cost. From 1 January 2027 the standard rate applies, working out near $2.15 and $10.65. Google lists API prices in United States dollars, so what an Australian team actually pays moves with the exchange rate. The figures here use the Reserve Bank rate from 13 August.

Developers get the model in Google AI Studio, Android Studio and Antigravity, while enterprises get it through the Gemini Enterprise Agent Platform. Consumers will meet it as Spark, a feature inside the Gemini app rolling out across more than 160 countries including Australia. The European Economic Area, the United Kingdom and Switzerland are excluded from the Spark rollout for now.

Built for agents rather than chat

Flash class models are the ones that run in volume. A single automated workflow can make dozens of model calls per task, which is why the per million token price decides whether an agent is commercially viable at all. The AutomationBench result is the number to watch here, since it tests exactly that kind of multi step tool use.

For coding teams the distribution matters as much as the score. The same engine now sits inside Android Studio and Antigravity, Google’s agentic coding environment, so the model a developer prototypes with is the one that later runs the production workflow. That removes a class of surprises that used to appear when teams swapped models between development and deployment.

Google is pitching 3.7 Flash as the default engine for that work rather than a budget compromise, and the distribution list backs that up. Putting the same model into developer tools, the Enterprise Agent Platform and the consumer app means one engine serving three audiences at once. For businesses weighing up the two big ecosystems, our ChatGPT vs Gemini comparison covers how the platforms differ beyond raw benchmarks.

The price war moves down market

Google is not moving in a vacuum. OpenAI cut GPT 5.6 Luna pricing by 80 per cent in July, a move we covered in our Luna price cut report. This week it also published builder guidance showing Luna matching its previous flagship on a browsing benchmark, 84.04 against 84.36 per cent, at a small fraction of the cost, per OpenAI’s own guide.

The pattern is consistent across vendors. The smaller models of six months ago were compromises, while the current crop handles most routine automation work at a tenth of flagship cost. Gemini 3.7 Flash at half price until December is Google’s bid to win those workloads before the year closes, and the pricing calendar makes the intent plain.

The price war moves down market

Why Australian businesses should care

For Australian teams building automation, the sums just changed. At about $1.05 per million input tokens, a document heavy workflow that processes thousands of pages a month runs at a fraction of what the same job cost on a flagship model earlier this year. Support agents, invoice extraction and internal search all sit squarely in the workload class this model targets. As a concrete yardstick, a pipeline that reads 20 million input tokens a month lands near $21 at the introductory rate and about $43 once standard pricing starts.

The introductory window also gives a clean test period. Running a real workload on 3.7 Flash between now and December produces actual per task costs to compare against what you pay today, and the January price rise is already published, so the budget case can be settled with data rather than guesswork. Choosing between model families has become a budgeting exercise as much as a technical one, the kind of question our AI consulting work deals with weekly.

A deadline that does the deciding

Google puts Gemini’s user base at a billion people, and the Spark rollout is designed to push 3.7 Flash into daily use at that scale. The model behind a consumer feature and the model behind an invoice pipeline are now the same engine, priced to spread as widely as possible.

The date that matters is 1 January. Whatever the benchmarks say, the introductory price is the recruitment offer and the standard rate is the relationship. Australian businesses that start measuring now will know by Christmas whether the workhorse earns its place.

Scroll to Top