TL;DR
At Google I/O on May 19, 2026, Google announced Gemini 3.5, its next family of models combining frontier intelligence with action. The series opens with 3.5 Flash, which Google says rivals flagship models on coding and agentic benchmarks while running roughly 4x faster on output tokens than other frontier models and at less than half the cost. 3.5 Pro is in internal use and ships next month.
For CX leaders, 3.5 Flash is the most consequential model in the release because Flash-class models already carry the majority of production support traffic. A Flash that lands top-right on the Artificial Analysis Intelligence vs Speed chart changes the routing math.
What Changed vs Gemini 3.1
Five things are materially different versus Gemini 3.1.
Flash now beats 3.1 Pro on agentic and coding benchmarks. Google reports 3.5 Flash outperforms Gemini 3.1 Pro on Terminal-Bench 2.1 (76.2%), GDPval-AA (1656 Elo), and MCP Atlas (83.6%), and leads multimodal understanding at 84.2% on CharXiv Reasoning. In plain English: the cheap, fast tier of the new family is stronger than the expensive tier of the previous one on the work CX agents actually do.
4x faster output than other frontier models. Output tokens per second is roughly 4x competing frontier models. For synchronous chat, that is the difference between a human-feeling reply and a visible wait.
Around half the cost of rival frontier models. Google explicitly positions 3.5 Flash as completing work in a fraction of the time and often at less than half the cost. Combined with the speed gains, this reshapes the economics of routing higher-tier traffic.
Built for long-horizon agentic work via Antigravity. Coupled with Google's updated Antigravity harness, 3.5 Flash deploys collaborative subagents that plan, build, and iterate on multi-step workflows. The launch ships with worked examples (legacy codebase migrations to Next.js, multi-agent game development, parallel UX generation) that translate cleanly to CX back-office operations: bulk refund sweeps, KB rewrites, end-of-day reconciliations.
Default model for Gemini app and AI Mode in Search. 3.5 Flash is now the default model behind the Gemini app and AI Mode in Google Search globally. It also powers the new Gemini Spark personal agent. The signal: Google is putting its highest-volume consumer surface on Flash, which means the production hardening is real.
Industry Reception
Google published partner case studies alongside the launch. The pattern is consistent.
Shopify is running 3.5 Flash subagents in parallel to analyse complex data over long horizons for merchant growth forecasts at global scale. Macquarie Bank is piloting 3.5 Flash to accelerate customer onboarding by reasoning over 100+ page documents and making low-latency recommendations. Salesforce is integrating 3.5 Flash into Agentforce for multi-turn enterprise tool calling across subagents. Ramp uses it for OCR on complex invoices combined with reasoning over historical patterns. Xero is deploying agents that autonomously manage multi-week workflows like 1099 tax form preparation. Databricks uses agentic workflows to monitor real-time data, reason across massive datasets, and propose fixes.
Two themes recur: long-horizon reliability, and document-heavy multimodal reasoning at production cost.
What This Means for CX Teams
The practical implications are sharper than for Opus 4.8, because Flash-class models carry the bulk of tier-1 traffic.
Swap the model ID and measure. If you are already routing chat traffic to Gemini 3.1 Flash or 3.1 Pro, run 3.5 Flash in parallel on the same case mix. Same surface, better benchmarks, faster output, lower cost. The migration is a string change.
Consider Flash for work currently sent to Pro-class models. Where 3.1 Pro was your escalation tier for complex reasoning, 3.5 Flash is now in the same range on agentic and coding benchmarks at a fraction of the cost and latency. Validate on your hardest cases before reshaping routes.
Use Antigravity-style subagent patterns for back-office operations. Multi-week, multi-step workflows (refund reconciliations, supplier verification, bulk KB updates, post-incident audits) are where the long-horizon capability pays back fastest. Pilot on internal operations before customer-facing flows.
Lean into multimodal for document-heavy support. Insurance claims, returns with photos, complex invoices, KYC packs. The CharXiv leadership and the Ramp / Macquarie patterns translate directly into CX flows where customers attach documents or images.
Do not retire your routing discipline. Flash gains do not eliminate the case for tiering. The right architecture is unchanged: route each contact to the cheapest model capable of resolving it at the required quality bar. 3.5 Flash just shifts where that line sits.
AI Readiness Score
How ready is your team for AI?
6 quick questions. Get a personalised score and action plan.
Try the AI Readiness Score1000+ agents deployed worldwide · 4.8 on G2
How Gemini 3.5 Flash vs Gemini 3.1 Stacks Up
Terminal-Bench 2.1: 76.2% on 3.5 Flash, beating 3.1 Pro. GDPval-AA: 1656 Elo on 3.5 Flash, beating 3.1 Pro. MCP Atlas (agentic tool use): 83.6% on 3.5 Flash. CharXiv Reasoning (multimodal): 84.2%, leading the comparison set. Output speed: roughly 4x competing frontier models. Cost: roughly half of rival frontier models on equivalent tasks. Default surface: now powers Gemini app and AI Mode in Search globally.
The Honest Caveats
Two things to watch.
Benchmark leadership is not the same as production reliability. Google's partner case studies are encouraging, but every CX team should run their own A/B on real case mix before flipping default routes.
3.5 Pro is not out yet. Google says it is being used internally and rolls out next month. If your escalation tier depends on Pro-class judgement, plan for two migrations, not one.
What to Do This Week
Three concrete moves.
Add gemini-3.5-flash to your routing matrix. Mirror a slice of your current Gemini 3.1 traffic to it. Measure CSAT, autonomy, latency, and cost per resolution over two weeks.
Identify one long-horizon back-office workflow. Pick a multi-step internal operation (refund sweeps, supplier verification, KB rewrites). Pilot 3.5 Flash with subagent patterns through Antigravity or Gemini Enterprise. Measure throughput and accuracy.
Audit your multimodal flows. Anywhere customers attach documents or images, run the current model and 3.5 Flash side by side. Track extraction accuracy, downstream resolution, and human-touch rate.
How Certainly Helps CX Teams Navigate Model Releases
Choosing between Gemini 3.5 Flash, Claude Opus 4.8, GPT-5.5, and the rest is not a one-off procurement call. It is a continuous routing decision. Certainly has spent the last 10 years helping global brands like Rockwool, Entain, McDonalds, and Carlsberg deploy production bots and agents, and our platform routes contacts across multiple frontier models with the right model on the right case.
If you want a structured read on what to migrate, what to leave alone, and where the ROI actually shows up, book a live demo, generate a business case, or talk to our team. We will bring 10 years of production deployment experience to the conversation.
Try Gemini 3.5 Flash directly in the Gemini app, Google AI Studio, or Gemini Enterprise.