Felo AI Adds Gemini 3.6 Flash — Free Access to Google's New Workhorse Model
Google just shipped Gemini 3.6 Flash with lower pricing and stronger agentic performance. Felo AI now supports it — try it free, no credit card needed.
Google shipped three new Gemini models on July 21. Felo AI already runs the one that matters most for everyday work: Gemini 3.6 Flash.
It's live now in Felo Search and Felo Agent. Free to try, like everything else on the platform.
What Is Gemini 3.6 Flash?
Flash has always been Google's speed play — the model built for volume, not just raw power. 3.6 keeps that job description but gets noticeably better at it.
It sits between two siblings Google released the same day: Gemini 3.5 Flash-Lite, a cheaper option for simple tasks, and Gemini 3.5 Flash Cyber, a narrow security-focused variant. 3.6 Flash is the one built to be a daily driver — coding, research, multi-step agent work, all at a lower cost than its predecessor.
Output pricing dropped to $7.50 per million tokens, down from $9.00 for 3.5 Flash. Input stayed at $1.50.
That's not a headline number. It's the kind of quiet price cut that adds up fast once you're running thousands of searches a day.
Why It Matters
It does more with less. Per the Artificial Analysis Index, 3.6 Flash uses 17% fewer output tokens than 3.5 Flash to reach the same answer. On some coding benchmarks, that drop hits 65%. Fewer tokens means faster responses and lower cost — without asking you to write shorter prompts or settle for shorter answers.
Coding and agent work jumped. DeepSWE scores climbed from 37% to 49%. MLE Bench went from 49.7% to 63.9%. Those aren't small revisions — they're the difference between a model that needs hand-holding through a multi-step task and one that just finishes it.
Computer use is built in. 3.6 Flash ships with native computer-use as a client-side tool, and its OSWorld-Verified score jumped from 78.4% to 83.0%. That's the benchmark for a model actually operating a screen — clicking, typing, navigating — not just describing what it would do.
Knowledge work is sharper. GDPval-AA v2, which measures real professional tasks like document analysis and research synthesis, rose from 1349 to 1421.
Google says the model takes fewer reasoning steps and fewer tool calls to finish the same workflows. In practice, that's the gap between waiting on an agent and actually trusting it.
How It Works on Felo AI
In Felo Search, Gemini 3.6 Flash is now selectable from the model picker. Ask a question, pick 3.6 Flash, get an answer built from live, multi-language search results — not a static training cutoff.
In Felo Agent, it's the model handling the harder jobs: multi-step research, document synthesis, and workflows that used to require a slower, pricier model to get right. Because it's cheaper and faster than 3.5 Flash, agent runs that once felt sluggish now move at a pace that's actually usable.
Both surfaces are free. No credit card, no waitlist.
Who This Is For
If you're running research loops, generating reports from search results, or building agent workflows that chain multiple steps together, this is the model to reach for first. It's not the biggest Gemini model Google makes — that's still Pro. But for the volume of work most people and teams actually do day to day, 3.6 Flash is now the better trade: close to Pro-level reasoning on agentic tasks, at Flash-level cost.
Google is already teasing Gemini 4. Until then, 3.6 Flash is the sharpest tool in the lineup for anyone doing real work, not just testing prompts.