Gemini 3.5 Flash release + agentic push + Spark/Omni/Antigravity at I/O 2026
Key Questions
How does Gemini 3.5 Flash compare to flagship models on benchmarks?
Gemini 3.5 Flash achieves scores close to flagships such as Terminal-Bench 76.2% and MCP Atlas 83.6% at roughly one-third the cost. It also leads on CharXiv and MMMU-Pro.
What speed improvements does Gemini 3.5 Flash offer?
The model runs 4x faster than many rival AI systems. This makes it suitable for high-volume enterprise and real-time applications.
What is Spark in Google's agent ecosystem?
Spark is a 24/7 autonomous agent integrated with Workspace and Gmail. It partners with external services to handle ongoing tasks.
What capabilities does Gemini Omni provide?
Gemini Omni excels at multimodal video and physics tasks with a 2,887 ELO rating and 77.1% on ARC-AGI-2. It supports creation and editing of video content.
What is Antigravity 2.0 and its key features?
Antigravity 2.0 is Google's standalone agent platform offering CLI, SDK, managed execution, and enterprise support. It was launched at I/O 2026.
When will the Gemini CLI be deprecated?
The Gemini CLI will stop working on June 18, 2026. Users are encouraged to migrate to newer agent platforms like Antigravity.
How many users does Gemini currently have?
Gemini has reached 900 million users with refreshed app UI and thinking features. Enterprise customers report over $1 billion in annual savings.
How does Gemini 3.5 Flash position itself against GPT-5.5 and Claude 4.8?
It competes directly on performance while emphasizing lower cost and speed. Google positions it as a practical alternative for agentic and enterprise workloads.
Gemini 3.5 Flash rivals flagships (Terminal-Bench 2.1 76.2%, MCP Atlas 83.6%, CharXiv 84.2%, MMMU-Pro 84%, AA Index 55 at 1/3 cost); 4x faster; Spark 24/7 autonomous agent with Workspace/Gmail + partners; Omni multimodal video/physics (2,887 ELO, 77.1% ARC-AGI-2); Antigravity 2.0 agent platform CLI/SDK; Android CLI for agentic coding; CLI deprecation June 18 2026; Kaggle benchmarks added; $1B enterprise savings claim; 900M users; app UI/Thinking refresh. Competes vs GPT-5.5/Claude 4.8.