Google Cloud cuts Gemini 1.5 prices 25 percent and makes 2 million token context generally available on Vertex AI

Google Cloud on Tuesday made a two million token context window for Gemini 1.5 Pro and Flash generally available on Vertex AI, and cut list prices for the models by up to 25 percent across supported regions. The update, effective today in North America, Europe, and parts of Asia Pacific, is framed as a cost and scale play for customers building retrieval augmented generation and analytics over large files.

Google said the changes aim to convert pilots into production as year end procurement cycles and new compliance requirements arrive. The EU AI Act begins phased obligations later this quarter for high risk systems, which has pushed buyers to seek stable pricing and data controls.

Pricing and performance

Google said list prices on Vertex AI for Gemini 1.5 Pro are reduced by 25 percent for both input and output tokens, while Gemini 1.5 Flash drops 20 percent. Batch inference on Vertex AI is priced below online serving, with Google citing an average 30 percent delta for scheduled jobs. The company did not change existing committed use discounts, which can lower spend by up to 30 percent on 12 month terms.

In internal tests shared with reporters, Gemini 1.5 Flash delivered a median end to first token latency of 550 milliseconds at a 16,000 token prompt, which Google said is 15 to 20 percent faster than comparable small multimodal models used for chat and tooling. Google reported that the two million token context improved retrieval on long document tasks, citing a nine point gain on the LongBench summary F1 over the preview build. External verification was not immediately available.

Enterprise controls and compliance

Vertex AI now offers data residency for Germany and France with audit logging in Cloud Logging, plus no training on customer prompts by default under Google’s existing data processing terms. Google added a 99.9 percent availability SLA for Gemini 1.5 endpoints and expanded regional coverage to 11 locations including Frankfurt, Paris, London, Iowa, Oregon, Tokyo, and Singapore.

Google Cloud CEO Thomas Kurian said the company is aligning model economics with broader cloud commitments and BigQuery and Workspace integrations. Synergy Research Group estimates Google Cloud held about 13 percent share of global cloud infrastructure services in the second quarter of 2026, behind AWS at roughly 31 percent and Microsoft at about 25 percent.

The move extends a price and latency contest among model platforms that has intensified on Tuesdays as vendors stack announcements. Lower costs for long context jobs pressure rival APIs from OpenAI and Anthropic and increase demand for high bandwidth accelerators, a dynamic that benefits suppliers like Nvidia with Blackwell systems and AMD with MI300 class parts now in hyperscale fleets.