Google Releases Gemini 3.6 Flash and 3.5 Flash-Lite for Agentic Workflows
Google introduced Gemini 3.6 Flash and Gemini 3.5 Flash-Lite, designed to lower latency and token usage in agentic workflows. Gemini 3.6 Flash reduces output tokens by 17% while costing $1.50/1M input and $7.50/1M output, and adds native client-side computer use capabilities.
Impact: High
Why it matters
Developers can immediately swap 3.5 Flash for 3.6 Flash or 3.5 Flash-Lite to reduce token costs and improve multi-step agent performance.
TL;DR
- 01Gemini 3.6 Flash delivers a 17% reduction in output token usage and costs $1.50/$7.50 per 1M tokens.
- 02Gemini 3.5 Flash-Lite achieves 350 tokens/sec at $0.30/$2.50 per 1M tokens with SWE-Bench Pro score of 54.2%.
- 03Computer use is now available natively as a client-side API tool across Gemini Flash models.
Key facts
- 3.6 Flash Input Price
- $1.50 / 1M tokens
- 3.6 Flash Output Price
- $7.50 / 1M tokens
- 3.5 Flash-Lite Speed
- 350 tokens/sec (self-reported)
- 3.5 Flash-Lite Input Price
- $0.30 / 1M tokens
- 3.5 Flash-Lite Output Price
- $2.50 / 1M tokens
- 3.6 Flash DeepSWE Benchmark
- 49% (self-reported)
Model Specifications and Pricing
Google's Gemini 3.6 Flash is priced at $1.50 per 1M input tokens and $7.50 per 1M output tokens. Gemini 3.5 Flash-Lite operates at $0.30 per 1M input tokens and $2.50 per 1M output tokens, achieving an execution speed of 350 output tokens per second.
Benchmark Improvements and Token Efficiency
- Gemini 3.6 Flash achieves 49% accuracy on DeepSWE (up from 37% in 3.5 Flash) and 63.9% on MLE Bench.
- On OSWorld-Verified, 3.6 Flash scores 83.0% with built-in client-side computer use support.
- Gemini 3.5 Flash-Lite reaches 54% on Terminal-Bench 2.1, 72.2% on GDM-MRCR v2, and 54.2% on SWE-Bench Pro.
Specialized Domain Models
Google also announced Gemini 3.5 Flash Cyber, fine-tuned specifically for vulnerability detection and automated patching within the CodeMender framework.
Try it in 2 minutes
curl https://generativelanguage.googleapis.com/v1beta/models/gemini-3.6-flash:generateContent?key=$GEMINI_API_KEYbash
✓ When to use
- Use Gemini 3.6 Flash as the execution engine in multi-agent orchestration to cut token costs by up to 17%
- Use Gemini 3.5 Flash-Lite for high-throughput sub-agent tasks like search, document parsing, and visual web design generation
✕ When NOT to use
- Avoid using Flash-Lite as the primary planning or reasoning master agent for complex multi-step architecture design
What to do today
- Update Gemini SDK to test 3.6 Flash in agent execution pipelines to reduce output token consumption.
- Benchmark 3.5 Flash-Lite as a high-throughput worker for document parsing and agentic search.
What the community says
“Fable or gpt5.6 sol for planning. Gemini 3.6 Flash for executing. Wow, Google is onto something here.”
“This is a very fast model. I was already impressed by how fast 3.5 Flash was.”
Sources