Google delays Gemini 3.5 Pro, releases 3.5 Flash at I/O

Google unveiled the Gemini 3.5 family at Google I/O 2026 and made the lighter-weight Gemini 3.5 Flash available immediately, according to Google's official blog post titled "Gemini 3.5: frontier intelligence with action" (May 19, 2026). Reporting by Business Insider documents that CEO Sundar Pichai told the audience "I know you can't wait to get your hands on it, Give us until next month to get it to you," and that some attendees groaned when Gemini 3.5 Pro was not released at the event. CNBC and Google's blog state that Gemini 3.5 Pro is being used internally and is expected to roll out next month.

Per Google's blog, Gemini 3.5 Flash is optimized for agentic workflows and coding, and Google presents benchmark claims where it outperforms Gemini 3.1 Pro on metrics reported by Google (Terminal-Bench 2.1: 76.2%, GDPval-AA: 1656 Elo, MCP Atlas: 83.6%) and offers much higher output throughput. Google's blog and Ars Technica reporting describe 3.5 Flash as delivering substantially higher tokens-per-second output (Google claims it produces outputs at roughly 4x the token rate of some frontier models) while matching or exceeding key agentic and coding metrics. The Google Cloud blog lists additional product announcements tied to the 3.5 family, including Gemini Omni, Gemini Spark, Google Antigravity, and integration points for developers via the Gemini API and Agent Platform.

Editorial analysis: Companies deploying agentic AI at scale face cost and latency pressures; public reporting on 3.5 Flash frames it as an efficiency play that aims to make longer-running agentic tasks more economically viable. Multiple outlets (CNBC, Ars Technica, TechCrunch summaries) emphasize that major AI vendors are pursuing model-efficiency improvements to reduce operational costs for persistent agents and coding workloads.

Editorial analysis: For practitioners, the immediate availability of a high-throughput model variant like 3.5 Flash matters because it can change the cost-performance tradeoffs for agentic systems, long-horizon orchestration, and real-time developer tooling. Reporting by Google and coverage across CNBC and Ars Technica indicate Google is integrating 3.5 Flash broadly, into the Gemini app, AI Mode in Search, Google Cloud developer surfaces, and enterprise offerings, which increases the practical touchpoints where engineers will encounter the model.

Editorial analysis: Observers and practitioners should monitor:

• •independent benchmark replications of Google's claims for 3.5 Flash and 3.5 Pro
• •latency and cost figures from cloud pricing and enterprise tiers once 3.5 Pro is broadly available
• •how the Agent Platform and Google Antigravity tooling surface support for multi-step agent orchestration and observability. Also track security and safety evaluations, since Google says it strengthened defenses for 3.5 Flash (reporting summarized in CNBC)

• •Google's blog post "Gemini 3.5: frontier intelligence with action" lays out feature claims and benchmark numbers for 3.5 Flash and notes 3.5 Pro is used internally and slated for next-month rollout (Google blog, May 19, 2026).
• •The Google Cloud blog summarizes I/O announcements including Gemini Omni, Gemini Spark, and managed-agent APIs (Google Cloud blog, May 20, 2026).
• •Business Insider reports on audience reaction and quotes Sundar Pichai's onstage comments (Business Insider, May 19, 2026).
• •CNBC and Ars Technica provide independent coverage of the product positioning, efficiency claims, and enterprise implications (CNBC, Ars Technica, May 19, 2026).

Editorial analysis: Overall, the event advances Google's agentic stack by shipping an efficiency-optimized frontier model now while holding back the heavier 3.5 Pro variant until next month; practitioners should treat today's 3.5 Flash availability as an actionable option for building higher-throughput agentic and coding features, while awaiting 3.5 Pro for comparisons on raw capability and cost tradeoffs.