Google just released Gemini 3.5 Flash.
And honestly:
This is the first Google model in a while that genuinely feels competitive in modern coding workflows instead of just benchmark charts.
The surprising part?
This isn’t even the Pro model.
Historically, Google’s:
- Flash models
- Pro models
- Ultra models
have existed in a pretty predictable hierarchy.
Flash was usually:
- faster
- cheaper
- lighter
while Pro was the serious frontier model.
But Gemini 3.5 Flash changes that perception a bit.
This model already feels:
- highly agentic
- extremely fast
- multimodal
- visually creative
- unusually strong at front-end generation
And in some coding evaluations, it even appears to outperform Gemini 3.1 Pro.
That’s a pretty significant jump.
Gemini 3.5 Flash Feels Built for Agentic Workflows
One thing becomes obvious almost immediately when testing Gemini 3.5 Flash:
This model feels optimized for execution speed.
Not just token streaming speed.
Workflow speed.
It plans quickly. It iterates quickly. It generates large outputs quickly. It reacts quickly.
Even multi-thousand-line generations complete surprisingly fast.
Google heavily emphasized:
- responsiveness
- throughput
- inference speed
during launch. And honestly, it shows.
This feels much more aligned with:
- Codex-style workflows
- autonomous coding agents
- interactive development environments
- modern agentic runtimes
than older chatbot-style AI systems.





































































//Take Command of your code.
Ship 10x faster with the same team, less time, and your coding taste. Install, sign in, and start coding.
The Benchmarks Are Actually Interesting This Time
Normally benchmark slides don’t mean much.
But Gemini 3.5 Flash’s results are surprisingly hard to ignore.
Google showed the model outperforming:
- Gemini 3.1 Pro
- several coding benchmarks
- and even Claude Opus 4.7 in some agentic coding evaluations
particularly around:
- terminal coding workflows
- interactive orchestration
- harness-based execution
And again: this is still the Flash model.
Not Gemini 3.5 Pro.
That’s what makes the release feel important.
Because if the improvements scale upward into Pro: Google might finally have a genuinely top-tier coding model ecosystem again.

Command Code Already Supports Gemini 3.5 Flash
Command Code added support for Gemini 3.5 Flash in:
1v0.26.11This model feels particularly strong inside agentic coding workflows.
Especially because Gemini 3.5 Flash benefits heavily from:
- orchestration
- parallel execution
- long-context handling
- runtime coordination
- multimodal tooling
Inside Command Code, use Gemini 3.5 Flash for:
- parallel agentic execution
- multimodal coding workflows
- 1M token context handling
- orchestrated coding sessions
Run the /model command in Command Code session to switch between models.





































































//Take Command of your code.
Ship 10x faster with the same team, less time, and your coding taste. Install, sign in, and start coding.
The Speed Might Be the Most Important Part
The thing that keeps standing out with Gemini 3.5 Flash is:
how fast it feels.
Not benchmark fast. Interactive fast.
The model:
- plans rapidly
- generates quickly
- iterates quickly
- handles large outputs well
- recovers quickly during execution
That changes the feeling of using it.
And in agentic coding systems: speed matters a lot more than people realize.
Because autonomous workflows increasingly depend on:
- rapid retries
- iterative planning
- tool execution loops
- orchestration cycles
Latency compounds quickly in those environments.
Gemini 3.5 Flash feels optimized for exactly that style of workflow.
The Most Interesting Part?
This Still Isn’t Gemini Pro
This is probably the biggest takeaway.
Gemini 3.5 Flash already feels:
- highly capable
- visually creative
- strongly agentic
- multimodal
- orchestration-friendly
And this still isn’t:
Gemini 3.5 Pro.
Which makes the upcoming Pro release much more interesting.
Because if Google scales these improvements upward: they may finally have a genuinely top-tier coding ecosystem again.
Especially as coding quality increasingly becomes about:
- orchestration
- execution
- multimodal workflows
- runtime systems
instead of pure benchmark scores.
Try It in Command Code
1npm i -g command-codeSign up for Command Code. Install it, run cmd, write some code using Gemini 3.5 Flash.
