Gemini 3.7 Flash Price: Half the Cost of 3.6 Flash for Agentic Coding

On August 13, 2026, Google released Gemini 3.7 Flash, and the number that matters most for heavy AI users is the price: $0.75 per million input tokens and $3.75 per million output tokens. That is half the original launch cost of Gemini 3.6 Flash, and it lands inside Google’s most capable coding and agent model to date. For anyone paying $300 or more a month across AI tools, the gemini flash price just became the new reference point for cheap frontier workhorse tokens.

The launch comes just three weeks after Gemini 3.6 Flash, an unusually fast cadence that Google attributes to developer feedback and algorithmic gains. The message is deliberate: Google is fighting hard on price and performance in the workhorse tier, the exact segment where heavy users burn most of their agentic token spend.

Gemini 3.7 Flash Price: How the Numbers Stack Up

The headline gemini flash price is a straight line down from the previous generation. Where Gemini 3.6 Flash launched at $1.50 per million input tokens and $7.50 per million output tokens, Gemini 3.7 Flash cuts both rates in half. Google is holding that introductory pricing through the end of 2026.

Model	Input per MTok	Output per MTok	Rough 1M in + 1M out
Gemini 3.7 Flash	$0.75	$3.75	~$4.50
Gemini 3.6 Flash	$1.50	$7.50	~$9.00
Claude Sonnet-class	varies	varies	higher

For an agentic coding workload that is half input and half output, a million tokens of each now costs roughly $4.50 on 3.7 Flash, down from about $9.00 on 3.6 Flash. Double that with a typical agent’s output-heavy mix and the savings compound, because agent loops generate far more output rows than they consume in prompts. Price the task rather than the token, and 3.7 Flash is one of the cheapest ways to run a coding agent that does not routinely stall.

Frontier Coding Without the Frontier Bill

The meaningful part of this release is that the price cut did not come with a capability cut. Gemini 3.7 Flash posts a 43.6% on FrontierCode 1.1 Main, up from 34.4% on 3.6 Flash, and 65.3% on DeepSWE v1.1, up from 49.0%. On WebDev Arena it holds an Elo of 1588 against 1538 for 3.6 Flash. That is the profile of a genuine workhorse: strong first-pass code, better debugging and issue resolution, and fewer retries across an agent session.

Many heavy users route a large share of their agentic work through a fast, cheaper model and reserve flagship models for the hardest stumps. Gemini 3.7 Flash makes that routing decision easier, because it closes much of the quality gap that used to force everything onto a flagship lane. For repeated knowledge-work tasks, document processing, and multi-step business workflows, Google reports strong gains on GDP.pdf (34.0% vs 22.0%) and AutomationBench (30.4% vs 17.0%).

What Gemini 3.7 Flash Means for Gemini Spark Subscribers

There is a second story hiding in this release for subscription-based users. Google is switching Gemini Spark, its always-on personal agent, to Gemini 3.7 Flash for AI Pro and Ultra subscribers. Spark is the always-on agent that runs 24/7 while you work, acting in Google apps and Workspace. Moving it onto 3.7 Flash means better tool use, more accurate multi-step workflows, and cheaper inference behind the scenes.

For a heavy user already paying for AI Pro or Ultra, this is a meaningful upgrade to the included value you get from your subscription, rather than an extra cost. If you were holding a subscription mostly for the agent, the 3.7 Flash upgrade strengthens that calculation. If you are not subscribed and route agents through the API, the per-token economics now favor testing 3.7 Flash as your default cheap lane for everything that does not need a flagship.

Access and Practical Routing for Heavy Users

Gemini 3.7 Flash is available today through the Gemini API, Google AI Studio, and Android Studio, plus Antigravity for agent-first workflows and the Gemini Enterprise Agent Platform. That distribution matters for quota routing, because you can now run a coding agent against one of the cheapest frontier-tier models in the API without switching harnesses.

The honest caveat is that 3.7 Flash is still a fast-tier model, not a replacement for the deepest reasoning lanes. It is equipped for a large share of production agent work, but for long-horizon architecture surgery or the most complex debugging, you will still want a flagship model in the rotation. The practical play is to route the high-volume, output-heavy work to 3.7 Flash and keep the rare, hardest tasks on a premium lane.

The Pattern to Watch

The bigger signal for heavy AI users is the trajectory. Gemini 3.7 Flash is the second workhorse tier this month to ship a headline price cut, after OpenAI and Microsoft moved similar models within weeks of each other. The workhorse tier is becoming the battleground where providers compete hardest on cost per million tokens, because that is where agentic spend actually flows.

For budgeting, that shifts the smart default: price the task, not the token, and benchmark your real agent workloads on the cheap workhorse lane before committing long-term spend. Keep one flagship lane for the hardest problems, but route the high-volume agentic work onto the model whose gemini flash price is now the lowest it has ever been.

The Takeaway for Heavy AI Users

Gemini 3.7 Flash is a rare combination: a substantial capability jump plus a 50% price cut on the previous generation. For heavy AI users, the immediate action is to route high-volume, output-heavy agentic coding onto 3.7 Flash, benchmark it against your current cheap lane, and measure the cost per completed task rather than the per-token list price. For AI Pro and Ultra subscribers, the Gemini Spark upgrade to 3.7 Flash is a free improvement to value you already pay for. The workhorse pricing war is the most important cost trend of the quarter, and Google just fired a round that resets the reference price.