Google Halves the Price of Gemini 3.7 Flash: Task Volume Gets Recalculated
On 13 August, Google introduced Gemini 3.7 Flash. It is an update to the workhorse class of models used for code and AI agents. The previous version shipped three weeks earlier.
Where has the cheap class improved?
Through the end of 2026, the price is $0.75 per 1M input tokens. Output runs $3.75 per 1M. That is half the launch price of Gemini 3.6 Flash.
Google's own benchmark numbers: FrontierCode 1.1 Main — 43.6% against 34.4%, DeepSWE v1.1 — 65.3% against 49%. The Elo rating in WebDev Arena — 1588 against 1538. On business processes in AutomationBench — 30.4% against 17%.
Why redo the budget for high-volume work this quarter?
Document processing, code generation and support agents all run on the cheap class of models. Input costs half of what it did, so the same monthly volume comes out cheaper. The gain on AutomationBench points to fewer manual saves after an agent.
The math has to be run on real data. Take a month of actual traffic and run a sample of 200–300 tasks through both versions. Compare the share of tasks completed without human intervention. The inference bill and model switch calculators cover both sides of the equation.
What is still unknown?
The context limit, pricing after 31 December 2026, API and payment availability for Russian legal entities, and how long version 3.6 will be supported.
Let’s discuss your project?
Tell us about your process — we’ll suggest where AI pays off fastest.
Related articles
OpenAI Ships GPT-6 Astra: 63% Lower Cost Per Task
OpenAI unveiled GPT-6 Astra: a 12-point lead on the domain-specific benchmark and a claimed 63% lower price per task. Agent budgets are being recalculated.
Z.ai Opens GLM-5.3 for Coding and Agent Work
Z.ai has released GLM-5.3 with open weights. Per-token pricing for the new model has not been published yet, so budgets have to be built on the previous version's rate.