Google slashes Gemini 1.5 Flash prices up to 78%

Google is turning up the heat in the AI pricing wars.

Fresh off taking the top spot on the LMSYS leaderboard with an experimental Gemini 1.5 Pro build, Google just took an axe to the API costs of its workhorse lightweight model, Gemini 1.5 Flash, alongside a batch of major platform upgrades.

Slashing prices by over 70%

Starting August 12, Google is cutting Gemini 1.5 Flash input pricing by 78%, dropping it to just $0.075 per 1 million tokens for prompts under 128K tokens.

Output pricing is getting a similar treatment, falling 71% down to $0.30 per 1 million tokens.

These cuts also cascade down to prompts exceeding 128K tokens and context caching rates. When combined with long context caching, running bulk summarization or data extraction jobs just became dirt cheap.

On top of the discount, text tuning for Gemini 1.5 Flash is now officially rolled out to all developers via the Gemini API and Google AI Studio. This lets teams bake task-specific logic directly into custom base weights, trimming down system prompt lengths, latency, and recurring token costs.

Native PDF vision and a snappier Studio

Language support across both 1.5 Pro and Flash has expanded to over 100 languages, which Google says should eliminate unexpected language block finish errors in the API.

Document processing also got a big upgrade. The Gemini API and AI Studio now handle PDF understanding through combined text and vision, letting the model natively process charts, diagrams, and figures instead of scraping plain text.

Google Workspace accounts can now jump straight into Google AI Studio by default without extra setup, opening seamless access to millions of corporate users.

Meanwhile, the AI Studio web interface itself saw roughly a 50% drop in loading times, added drag-and-drop image uploads, revamped keyboard shortcuts, and introduced prompt suggestions alongside an overhauled API reference.

Why it matters

For developers and startups, Gemini 1.5 Flash just became an aggressive default for background pipelines. At 7.5 cents per million input tokens, use cases that were previously cost-prohibitive—like parsing massive archives, continuous customer ticket analysis, or categorization—are now trivial line items.

Opening up fine-tuning on Flash is another sharp move. Shifting long instructions out of the context window and into model weights cuts latency and saves money on every single API hit.

This is a direct shot at OpenAI's GPT-4o mini and Anthropic's Claude 3 Haiku. While frontier models grab the headlines, lightweight utility models win the bulk API volume. Google is flexing its infrastructure scale to undercut competitors and keep developers locked into its ecosystem.

Native multimodal PDF parsing is also a huge quality-of-life win, though developers will still want to stress-test complex financial tables against traditional OCR pipelines. Either way, at these rates, Flash makes a compelling case for developers to rerun their cost-benefit math immediately.