Google releases token-efficient Gemini models, cuts agent costs up to 65%

Google DeepMind today announced three new proprietary AI models aimed at reducing costs for AI agents—autonomous systems that take actions based on AI reasoning—while improving speed and capability at scale. The three releases are Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and Gemini 3.5 Flash Cyber.

Pricing for the new models reflects an aggressive push into cost efficiency. Gemini 3.6 Flash is priced at $1.50 per million input tokens and $7.50 per million output tokens via API. Gemini 3.5 Flash-Lite is dramatically cheaper at $0.30 per million input tokens and $2.50 per million output tokens. Gemini 3.5 Flash Cyber's pricing was not disclosed. To contextualize these prices: the prior Gemini 3.5 Flash model cost $1.50/$9.00 per 1M tokens, Gemini 3.1 Pro Preview costs $2/$12 per 1M tokens, and Google's previous cost leader, Gemini 3.1 Flash-Lite, remains at $0.25/$1.50 per 1M tokens.

Google claims the new models deliver cost reductions of up to 65% on long-horizon engineering tasks—the kinds of multi-step AI reasoning workloads where token consumption compounds. The tradeoff with the existing Gemini 3.1 Flash-Lite is clear: while it is still the cheapest per token, Gemini 3.5 Flash-Lite runs 2× faster, giving enterprises that prioritize speed higher efficiency. Google also signaled that Gemini 3.5 Pro is on the way, though neither pricing nor a release date has been provided.

Google DeepMind's release of three new models reflects an industry-wide push to reduce the computational cost of running AI agents—software that autonomously take actions based on AI reasoning. The company positioned these models as "among its most token-efficient yet," signaling a direct competitive response to cost pressures in the inference (AI response generation) market. Tokens are the fundamental unit of computation for large language models; fewer tokens consumed means lower operational costs for customers.

The pricing architecture reveals a deliberate strategy: Gemini 3.5 Flash-Lite undercuts the prior Flash generation at a dramatically lower rate ($0.30/$2.50 vs. $1.50/$9.00), making it accessible for cost-sensitive workloads, while Gemini 3.6 Flash sits between them with the promise of 65% cost savings on long-horizon engineering tasks—the type of multi-step work where token efficiency matters most. Notably, Google's previous cost leader, Gemini 3.1 Flash-Lite, remains the absolute cheapest per token but is now 2× slower; this trade-off suggests Google is allowing customers to choose between ultra-low cost and acceptable speed rather than forcing one constraint.

The announced but unpriced arrival of Gemini 3.5 Pro suggests Google is also refreshing its higher-capability tier, though details remain sparse.

No comments yet. Be the first to share your thoughts!