Google’s Speed Play Changes the Model Selection Game

Google released Gemini 3.5 Flash as generally available on May 23, 2026, marketed as a frontier-level model delivering 4x the speed of comparable competitors. With pricing at $1.50 per 1M input tokens and $9 per 1M output tokens, supported by 1M token context windows and 76.2% performance on Terminal-Bench 2.1, the move signals a strategic pivot: speed and cost efficiency are now table-stakes for competing in the high-capability tier.

This isn’t just another model release. It’s a direct challenge to the prevailing assumption that frontier intelligence requires accepting latency penalties and premium pricing. For European builders evaluating closed-model infrastructure, the timing matters.

Why This Matters for Irish and European Enterprises

The EU’s AI Act creates compliance obligations around transparency, safety testing, and auditability for high-risk systems. Many Irish and European companies have defaulted to closed, auditable models from US providers to simplify compliance workflows. Gemini 3.5 Flash’s aggressive positioning on both speed and price reshapes the vendor evaluation matrix.

For customer-facing applications—customer support, real-time document processing, interactive analysis—latency directly affects user experience and operational cost. A model that delivers comparable capability at 4x speed means:

  • Lower infrastructure costs: Fewer parallel instances needed to handle throughput
  • Better user experience: Sub-100ms response times become feasible for more use cases
  • Reduced carbon footprint: Fewer token operations per transaction
  • Faster compliance iteration: Speed in model evaluation cycles accelerates safety testing timelines

The Broader Release Landscape

The May 2026 release schedule has been deliberately sparse. Labs that didn’t ship in April—a period when five different providers released models scoring above 50 capability points in a single month—are likely in development cycles for their own next-generation releases. Google’s move suggests the competitive window is narrowing: if you’re not shipping speed improvements alongside capability, you risk losing enterprise customers evaluating total cost of ownership.

Alibaba’s Qwen3 Coder Next (focused on code-scale reasoning) and MiniMax’s M2.5 Highspeed (optimized for latency-critical workloads) indicate the entire market is recognizing a shift. Specialized, fast models are becoming the default assumption.

What’s Still Unclear

  • Long-term pricing stability: Will $1.50/$9 hold, or is this an introductory rate subject to revision?
  • European data residency: Does Google’s infrastructure meet GDPR requirements for sensitive enterprise workloads in Ireland and the EU?
  • Comparative safety profiles: How does Gemini 3.5 Flash’s safety performance stack against Anthropic Claude and OpenAI’s offerings on HRAIS evaluation frameworks?
  • Open-model parity: With EuroLLM-22B and other EU-sovereignty alternatives emerging, how does the speed-cost advantage of proprietary models compare?

For Irish builders currently locked into slower or more expensive models, this release is a forcing function to re-evaluate. For those designing new applications, the speed-first paradigm is now the standard to beat.

Source: Google AI/Gemini announcements