AI Model Release Tracker

Google Gemini 3.5 Flash + Omni; new Flash-tier models; Gemini Robotics 2 and ER 2 released; 3.6 Flash limitations; Gemini 4 confirmed to replace 3.5 Pro; 3.6 Flash cost cuts and Flash Cyber model

Google Gemini 3.5 Flash + Omni; new Flash-tier models; Gemini Robotics 2 and ER 2 released; 3.6 Flash limitations; Gemini 4 confirmed to replace 3.5 Pro; 3.6 Flash cost cuts and Flash Cyber model

Key Questions

What new Gemini Flash models did Google release?

Google released Gemini 3.6 Flash, 3.5 Flash-Lite, and Flash Cyber alongside making 3.5 Flash the default model. These target efficiency, speed, and specialized tasks like bug hunting.

How does Gemini 3.6 Flash compare to previous versions on benchmarks?

Gemini 3.6 Flash shows a rare regression, scoring below 3.5 Flash on some benchmarks despite token efficiency gains of 17% and lower pricing at $7.50/M output tokens. It excels in agentic tasks and computer use but lags in coding.

What is Gemini Flash Cyber designed for?

Flash Cyber specializes in vulnerability hunting, identifying 55 V8 issues compared to Opus 4.6's 36. Access is gated due to dual-use risks.

Why did Google release 3.6 Flash before 3.5 Pro?

The strategy emphasizes immediate token reduction, lower costs, and strong performance in knowledge work and agentic benchmarks like DeepSWE and OSWorld while Pro development continues.

What are the pricing and speed details for 3.5 Flash-Lite?

3.5 Flash-Lite offers 350 tokens per second at $0.30/M input and $2.50/M output, focusing on extreme cost-performance for enterprise use.

How does Gemini 3.5 Flash perform against GPT-5.5?

It has become the default model and surpasses GPT-5.5 on agentic MCP Atlas benchmarks while delivering strong results on MLE Bench and OSWorld.

What limitations exist for the new Flash models?

Flash Cyber remains gated, Pro is still delayed, and some benchmarks reveal regressions in the 3.6 Flash variant despite efficiency improvements.

Which providers offer Gemini 3.6 Flash via API?

It is available through OpenRouter and other platforms with competitive pricing focused on token efficiency for long-horizon tasks.

Gemini 3.5 Flash now default, surpassing GPT-5.5 on agentic MCP Atlas. New Flash-tier models: 3.6 Flash (17% fewer tokens, 1M context) but shows regression. Major robotics releases: Gemini Robotics 2 for full-body humanoid control and ER 2 for multi-robot coordination, with whole-body control, dexterity, multi-robot collaboration, available on AI Studio and private preview. Google confirms Gemini 4 will replace the delayed Gemini 3.5 Pro, with leadership changes. New ARC-AGI benchmarks: 3.6 Flash scores 60.4% on ARC-AGI-2 and 91.2% on ARC-AGI-1. Recently, Gemini 3.6 Flash cuts output costs, and the restricted Flash Cyber model is available for developers.

Sources (9)
Updated Aug 10, 2026
What new Gemini Flash models did Google release? - AI Model Release Tracker | NBot | nbot.ai