Google Ships 3 Gemini Models, Flagship Pro Delayed
.png)
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Google launched Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber, framing them as workhorses for scaling AI agents at lower latency and cost.
- Google confirmed it has started its "most ambitious pre-training run yet" for Gemini 4, per product lead Logan Kilpatrick.
- Gemini 3.6 Flash records ~50% lower average time-per-task than its predecessor, with token usage cut by up to 65% on complex coding tasks.
- Gemini 3.5 Flash-Lite reaches 350 output tokens/sec and gains 11 Intelligence Index points over the prior generation, per Artificial Analysis benchmarks.
- Gemini 3.6 Flash jumped from #21 to #12 in the Frontend Code Arena (1537 points), ranking top-10 in reference-based design and content creation.
- Gemini 3.5 Pro remains unreleased, with Alphabet earnings the next day likely to field questions about Google DeepMind's position vs. OpenAI and Anthropic, per analyst Shakeel Hashim.
- Gemini 3.6 Flash's cost per task fell ~18% vs. 3.5 Flash, though 3.5 Flash-Lite's cost more than doubled vs. the 3.1 generation due to new token pricing.
Why it matters: With Alphabet earnings due the next day and analysts publicly flagging that Google DeepMind has fallen behind OpenAI and Anthropic, Google chose to ship efficiency-tuned models rather than a new flagship. The 50% time-per-task reductions and 65% coding token cuts reposition Google as competing on agent economics (cost-per-task, throughput) instead of raw intelligence benchmarks—an enterprise-friendly pivot that buys runway while 3.5 Pro stays in testing.
.png)
