Google Launches New Gemini Flash Models to Lower Enterprise AI Costs

As Alphabet prepares to report its second-quarter earnings this Wednesday, the company is attempting to pivot the narrative from delays to efficiency. The new model lineup, released on Tuesday, emphasizes lower price points and high-speed throughput for AI agents, a strategy Google executives believe will help companies manage their ballooning AI budgets.

Gemini 3.6 Flash and the Efficiency Push

The centerpiece of Tuesday’s release is Gemini 3.6 Flash, an upgrade to the company’s workhorse model. According to SiliconAngle, the new model uses up to 17% fewer output tokens than its predecessor while demonstrating stronger performance in coding and knowledge-work benchmarks. The model is priced at $1.50 per million input tokens and $7.50 per million output tokens.

Early adoption is already underway at specialized firms. Niko Grupen, head of applied research at Harvey AI Corp., noted that the model’s efficiency gains are tangible in daily operations.

“Compared to its predecessor, Gemini 3.6 Flash showed strong gains in performance on our benchmarks and was notably more efficient, completing tasks 12% faster on average.”

Niko Grupen, head of applied research at Harvey

Alongside the flagship of the Flash series, Google introduced Gemini 3.5 Flash-Lite. Positioned as the most cost-effective model in the 3.5 family, it is designed for high-throughput tasks like document processing and search. It is priced at a lower rate per million input tokens and $2.50 per million output tokens, according to thenewstack.io.

Gemini 3.5 Flash Cyber vs. Anthropic’s Mythos

Google’s most pointed move is the introduction of Gemini 3.5 Flash Cyber, a model explicitly tuned to identify and patch software vulnerabilities. The model serves as a direct competitor to Anthropic’s Mythos, which currently commands a premium price of $10 per million input tokens and $50 per million output tokens. As The Verge reported, Google claims its model matches frontier performance on the CyberGym benchmark while operating at a fraction of the cost.

However, the company is exercising caution regarding the potential for misuse. Because a tool capable of finding security flaws can also be used to exploit them, Google is restricting access to a limited pilot for governments and trusted partners. In tests conducted by Google DeepMind’s Big Sleep team, the model identified 55 unique security issues in the V8 JavaScript engine, including 10 vulnerabilities that no other model detected.

The Missing Flagship and Investor Scrutiny

Despite the expansion of the Flash lineup, the absence of the flagship Gemini 3.5 Pro remains a significant point of contention for investors. Originally promised for a June release after being announced at the I/O developer conference in May, the model is still in testing. Reuters notes that the delay stems from the model falling short of internal performance goals, particularly in coding capabilities.

This delay coincides with a period of heightened market skepticism. Alphabet shares have lagged behind other Magnificent Seven stocks, down about 9% since late April. Dave Wagner, a portfolio manager at Aptus Capital Advisors, summarized the investor mood:

“While Google is missing the boat on AI coding and that’s a very real growing concern …

Dave Wagner, portfolio manager at Aptus Capital Advisors

Looking forward, Google has attempted to reassure stakeholders by confirming that its most ambitious pre-training run yet for Gemini 4 is officially underway. Whether these efficiency-focused updates can satisfy shareholders during Wednesday’s earnings call remains the primary question, as the company grapples with capital expenditure guidance that reached between $180 billion and $190 billion in April.