Google slashes Gemini 1.5 Pro prices and drops smarter updates

If you build with AI, you probably deal with two constant headaches: massive API bills for long documents, and models that just won't stop yapping. Google wants to fix both.

Google rolled out updated production-ready versions of its flagship models: Gemini-1.5-Pro-002 and Gemini-1.5-Flash-002. Alongside them came an improved experimental lightweight model, Gemini-1.5-Flash-8B-Exp-0924. They are available now for free testing in Google AI Studio and the Gemini API, as well as on Vertex AI for enterprise teams.

Smarter, faster, and much cheaper

Both new models pack solid benchmark gains. Google reported a roughly 7% jump on MMLU-Pro and a massive 20% surge on both the MATH benchmark and HiddenMath, Google's internal competition math dataset. Vision understanding and Python code generation also improved between 2% and 7%.

Critically, Google tuned the models to be more concise. Default outputs for summarization, QA, and extraction tasks are now 5% to 20% shorter based on developer feedback. That means faster responses and lower token bills right out of the box.

Then comes the real aggressive move: slashing prices. Starting October 1, 2024, for prompts under 128K tokens, Gemini 1.5 Pro will see a 64% price cut on input tokens, 52% off on output tokens, and 64% off on context caching. Google also significantly boosted rate limits for paid tiers, raising 1.5 Flash to 2,000 RPM (up from 1,000) and 1.5 Pro to 1,000 RPM (up from 360). On top of that, safety filters are no longer applied by default, giving developers full control to configure guardrails for their specific apps.

Why it matters

Gemini 1.5 Pro's 2M context window has always been an engineering marvel, but processing full-length movies, massive codebases, and 1,000-page PDFs at scale was simply too expensive for many production budgets. A 64% price cut combined with shorter default answers and doubled rate limits makes long-context pipelines genuinely viable for daily operations. If you previously benched Gemini due to pricing or strict refusals, it is definitely worth testing again right now.

Industry-wide, Google is turning its custom silicon and data center muscle into a severe pricing squeeze against OpenAI and Anthropic. The frontier race is no longer just about reasoning evals; it is about serving efficiency and unit economics.

We think Google made the right call by stepping out of developers' way. Disabling mandatory default safety filters and reducing model chattiness shows a company listening to real-world deployment complaints instead of trying to overly sanitize API behavior.