Gemini 3.6 Flash Pricing, Benchmarks vs 3.5 Flash

Gemini 3.6 Flash release date and API access

Google announced Gemini 3.6 Flash on July 21, 2026, and developers could call it through the Gemini API on launch day. It is generally available rather than a preview, and its model ID is gemini-3.6-flash. The same launch post introduced two siblings, 3.5 Flash-Lite and 3.5 Flash Cyber.

Developers reach it through the Gemini API in Google AI Studio and Android Studio, or in Google Antigravity. Enterprises use the Gemini Enterprise Agent Platform or the Gemini Enterprise app, and everyone else can open it in the Gemini app. The Gemini launch models table sets the three launch models next to Google's own positioning.

Flash Cyber is not part of the general release. Google says it will go only to governments and trusted partners in a limited-access pilot, so most developers will not have access to it.

Pricing and cost per million tokens for Gemini 3.6 Flash

At launch, Gemini 3.6 Flash charged $1.50 for every million input tokens and $7.50 for every million output tokens, against the $9 that Gemini 3.5 Flash charged for output. The input price did not change.

Flash-Lite is the cheapest model in the Gemini launch models table on both input and output. These are the prices from the July 2026 launch. The Gemini API pricing page now also lists newer Flash models, 3.7 and 3.8, so confirm the current rate there before you size a budget.

17% fewer output tokens and what it means for cost per task

Two different 17% figures appear in coverage, and only one of them is a price. On the Artificial Analysis Index, Google reports a 17% drop in the output tokens 3.6 Flash needs compared with 3.5 Flash. Separately, the price per output token fell from $9 to $7.50, a cut of 16.7%.

The two changes stack, so the output bill for the same task can drop by more than either figure alone. Here is the arithmetic:

• Price per output token: $7.50 ÷ $9 = 0.833
• Output tokens per task: 17% fewer leaves 0.83 of the old count
• Output cost per task: 0.833 × 0.83 = 0.69, which is about 31% lower

This holds only if the 17% saving shows up on your own workload, because the figure was measured on the Artificial Analysis Index, not on your tasks. Input cost is unchanged, so the total bill for a prompt-heavy job falls by less than 31%.

How Gemini 3.6 Flash compares with 3.5 Flash on benchmarks

By the numbers in Google's launch post, 3.6 Flash needs fewer output tokens, reasoning steps and tool calls than 3.5 Flash, and it scores higher on coding, computer-use and knowledge-work benchmarks. The 17% token drop comes from the Artificial Analysis Index. The launch post lists the other scores, including the DeepSWE pair.

Google's post also says "up to 65%" for some benchmarks such as DeepSWE, without naming what that percentage measures. Gemini 3.6 Flash scores 49% on DeepSWE, so the 65% is not its score, and the post does not say what it is compared with.

These figures do not add up to a verdict on overall quality. What they support is narrower: fewer tokens, a lower output price and higher benchmark scores than 3.5 Flash.

DeepSWE score: 49% against 37% for 3.5 Flash

3.6 Flash scores 49% on DeepSWE against 37% for 3.5 Flash, a gain of 12 percentage points. DeepSWE is a benchmark built by Datacurve, and Google cites it as evidence of fewer unwanted code edits and fewer execution loops.

DeepSWE score, as reported by Google

3.6 Flash is the highlighted bar.

02040

49Gemini 3.6 Flash

37Gemini 3.5 Flash

Both scores come from Google's own post, and a single benchmark cannot stand in for every workload. A team that runs a coding agent should replay a few of its own tasks on both models before it switches.

Which to use: 3.6 Flash, 3.5 Flash-Lite or 3.5 Flash Cyber

Use 3.6 Flash by default for coding, multi-step agents and knowledge work, and use 3.5 Flash-Lite where price and speed on high-volume jobs matter most. Flash Cyber is not a real option for most teams, because Google is limiting it to governments and trusted partners.

A team already on 3.5 Flash has the simplest case. 3.6 Flash costs less for output, uses fewer output tokens and scores higher on DeepSWE, so it is the first model to test as a replacement. Flash-Lite fits high-throughput, low-latency jobs.

The same post says Gemini 4 pre-training has started and that 3.5 Pro was still in testing with partners at launch. To choose a model while Pro is missing from the API, see Gemini 3.5 Pro is still not in the Gemini API. Of the three launch models, Gemini 3.6 Flash is the default pick and Flash-Lite the cheaper step down.