What Is Gemini 3.7 Flash? Google's Coding Workhorse

Inside Gemini 3.7 Flash: The $0.75/$3.75 Workhorse That Shipped 3 Weeks After 3.6

Gemini 3.7 Flash is a natively multimodal reasoning model from Google DeepMind, released August 13, 2026, and positioned as the workhorse of the Gemini 3 family for coding and agents. It carries a 1M-token context window and an introductory price of $0.75 per 1M input tokens and $3.75 per 1M output tokens — half the original 3.6 Flash cost — good through the end of 2026. Google shipped it just three weeks after Gemini 3.6 Flash, crediting algorithmic improvements rather than a new base model. It is available now in the Gemini API, Google AI Studio, Antigravity, and Vertex AI.

Key Takeaways

• Released August 13, 2026, three weeks after Gemini 3.6 Flash (Google's launch blog).
• Introductory pricing is $0.75/1M input and $3.75/1M output tokens through end of 2026, described as 50% below 3.6 Flash.
• Context window is 1,048,576 tokens, with a 65,536-token max output, per Google's model documentation.
• Model ID is gemini-3.7-flash; it is built on Gemini 3.6 Flash with algorithmic upgrades, per the DeepMind model card.
• Coding scores jumped sharply over 3.6 Flash: FrontierCode 1.1 Main 43.6% vs 34.4%, DeepSWE v1.1 65.3% vs 49.0%.
• It powers Gemini Spark, Google's 24/7 personal agent, available to AI Pro and Ultra subscribers in 160+ countries.

What Is Gemini 3.7 Flash?

Gemini 3.7 Flash is the next iteration in Google's Gemini 3 series of high-throughput, multimodal reasoning models. Google DeepMind describes it as "our most intelligent workhorse model yet for coding and agents." The model is fully released, not a preview — it first surfaced in leaks on August 7, 2026, was spotted live on Vertex AI mid-morning on August 13, and got an official launch that same afternoon.

The Gemini Enterprise documentation frames its role plainly: 3.7 Flash "serves as the primary agentic workhorse in the Gemini 3 family, bridging the gap between deep-reasoning Pro models and high-throughput Flash-Lite models." In other words, it targets the volume tier where speed and cost matter more than peak intelligence.

Per the DeepMind model card, Gemini 3.7 Flash is based on Gemini 3.6 Flash, with the gains coming from "algorithmic improvements to its core reasoning foundation" rather than a fresh architecture. That is why Google could ship it three weeks after its predecessor.

Gemini 3.7 Flash at a Glance

Spec	Detail
Developer	Google DeepMind
Model ID	gemini-3.7-flash
Type	Multimodal reasoning model (Gemini 3 family)
Input modalities	Text, image, audio, video, PDF
Output	Text
Context window	1,048,576 tokens
Max output	65,536 tokens
Introductory price	$0.75 / 1M input, $3.75 / 1M output (through end of 2026)
Thinking levels	LOW, MEDIUM (default), HIGH (MINIMAL not supported)
Availability	Gemini API, AI Studio, Antigravity, Vertex AI, Gemini Enterprise, Spark
License	Proprietary (no open weights)
Base model	Gemini 3.6 Flash
Released	August 13, 2026

How Gemini 3.7 Flash Works / What Makes It Different

The headline mechanic Google emphasizes is what it calls "thinking more diligently." According to the launch blog, 3.7 Flash "puts in more effort into multi-step planning and tool calls," aiming for "more disciplined execution" with "less manual oversight and fewer retries."

The model exposes a Thinking Level control with three settings — LOW, MEDIUM, and HIGH — letting developers trade quality against cost and latency. Notably, the MINIMAL setting available on some models is disabled here; the docs state that requesting it returns a validation error.

Because 3.7 Flash is built directly on the 3.6 Flash foundation, its architecture, training data, and hardware profile inherit from that model card. The differentiation is in reasoning discipline and tool reliability, not raw scale. Google says the improvements arrived through "algorithmic innovations that we look forward to bringing to future models."

Source: @OfficialLoganK

On pricing, the anchor claim is simple: "50% lower price than 3.6 flash (through end of year)," in Logan Kilpatrick's words. Simon Willison flagged the structure as unusual, noting the introductory price is "scheduled to double in price on December 31, 2026" — a schedule the official launch posts confirm as end-of-year, though the exact 2027 rate is not formally published.

What You Can Do With Gemini 3.7 Flash

Google positions the model squarely at software engineering, agentic workflows, web development, and knowledge work in finance, law, and biosciences. The real-world use cases from launch day back that framing:

• Fast app and game generation. One developer built a playable Lunar Lander game in AI Studio in 54 seconds, with the prompt-to-game speed shown at 1x.
• Agentic coding loops. The model is available in Cline, Devin, and Google's own Antigravity agent IDE, where Max Weinbach reported "really fast" performance with "insanely high" usage limits.
• Multi-agent orchestration. One demo ran three 3.7 Flash agents together to autonomously train a robotics model from scratch, per the launch blog.
• Enterprise document work. Google cites gains on GDP.pdf (34.0% vs 22.0%) for processing complex documents, and on AutomationBench (30.4% vs 17.0%) for business workflows.

For teams already comparing Google's chat-tier models, our Gemini 3.6 Flash model page covers the predecessor this release builds on.

How Gemini 3.7 Flash Compares

Google's own model card benchmarks 3.7 Flash against its predecessor and three rivals. These are Google's launch figures unless noted, and independent replication remains thin.

Benchmark	Gemini 3.7 Flash	Gemini 3.6 Flash	Claude Sonnet 5	GPT-5.6 Terra
Input price ($/1M)	$0.75	$0.75	$2.00	$2.00
Output price ($/1M)	$3.75	$3.75	$10.00	$12.00
Intelligence Index	56	52	55	57
FrontierCode 1.1 Main	43.6%	34.4%	42.7%	41.3%
DeepSWE v1.1	65.3%	48.6%	53.8%	69.6%
Code Arena (Elo)	1588	1538	1541	1523
AutomationBench	30.4%	17.0%	10.7%	23.6%

The most-cited external validation came from Cognition, maker of Devin. Its own FrontierCode 1.1 benchmark landed 3.7 Flash at 56.3, "essentially tied with Claude Sonnet 5's 56.2," at less than half the cost. Bindu Reddy summarized the community read: "3x cheaper than Sonnet... just below Kimi K3 but above Terra." For a fuller picture of that mid-tier rival, see our Claude Sonnet 5 deep dive.

Not everyone was impressed. One benchmark tracker called the results "underwhelming," noting the model sits "below terra in a lot of benchmarks."

Availability: How to Access Gemini 3.7 Flash

Gemini 3.7 Flash is generally available now. Google lists these official distribution channels on the DeepMind model card:

• Gemini API via Google AI Studio — try it in the AI Studio playground.
• Google Antigravity — the agent-first developer environment.
• Vertex AI / Gemini Enterprise Agent Platform — with global and multi-region (us, eu) availability.
• Gemini App (Spark) — the model now powers Gemini Spark for AI Pro and Ultra subscribers across 160+ countries.
• Android Studio — cited in Google's launch materials.

Third-party agent tools including Cline and Devin added same-day support. There are no open weights; access runs through Google's API and partner integrations only.

What We Don't Know Yet

A reference page should be honest about the gaps. As of this writing:

• The post-2026 standard price is not officially published. Community posts predict the rate doubles on January 1, 2027, but there is no formal Google confirmation of the exact figure in the reporting so far.
• Benchmark gains are largely Google's own claims. Cognition and Box supplied independent evaluations, but broad third-party replication with published test settings is not yet available.
• Real-world quality beyond coding is mixed. One hands-on game-building test found 3.7 Flash faster but buggier than Kimi K3; another reported faster, cheaper task completion without declaring a winner.
• The fate of Gemini 3.5 Pro is unclear. One unconfirmed leak claimed it was "completely cancelled," and Google has not addressed the Pro line's timing.

Frequently Asked Questions

What is Gemini 3.7 Flash?

Gemini 3.7 Flash is a natively multimodal reasoning model from Google DeepMind, released August 13, 2026, positioned as the workhorse model in the Gemini 3 family for coding and agents. It has a 1M-token context window and introductory pricing of $0.75 per 1M input and $3.75 per 1M output tokens.

How much does Gemini 3.7 Flash cost?

Gemini 3.7 Flash costs $0.75 per 1M input tokens and $3.75 per 1M output tokens as an introductory price through the end of 2026. Google describes this as half the original 3.6 Flash cost per million tokens.

When was Gemini 3.7 Flash released?

Gemini 3.7 Flash was released on August 13, 2026, just three weeks after Gemini 3.6 Flash, according to Google's launch blog and official accounts.

Is Gemini 3.7 Flash open source?

No. Gemini 3.7 Flash is a proprietary Google model available through the Gemini API, Google AI Studio, Antigravity, Vertex AI, and Gemini Enterprise. No model weights have been released.

What is the context window of Gemini 3.7 Flash?

Gemini 3.7 Flash has a context window of 1,048,576 input tokens (about 1M) and a maximum output of 65,536 tokens, per Google's official model documentation.

How does Gemini 3.7 Flash compare to Claude Sonnet 5?

Gemini 3.7 Flash scored essentially tied with Claude Sonnet 5 on Cognition's FrontierCode 1.1 benchmark (56.3 vs 56.2) at less than half the cost. On Google's own model card, 3.7 Flash posts 43.6% on FrontierCode 1.1 Main versus Sonnet 5's 42.7%.

What can you do with Gemini 3.7 Flash?

Gemini 3.7 Flash is built for coding, debugging, agentic workflows, web development, and knowledge work across finance, law, and biosciences. It powers Gemini Spark and is available in agent tools like Antigravity, Cline, and Devin.

What to Watch Next

Three signals will define whether 3.7 Flash holds its lead. First, the January 1, 2027 pricing transition — whether Google confirms and how much the rate rises will decide if the current value case survives. Second, independent benchmark replication: watch for third-party evals with published settings that either confirm or deflate the Sonnet 5-parity claim. Third, the Gemini Pro line — the cancelled-3.5-Pro rumor and any Gemini 4 timeline will reshape where this Flash model sits in Google's stack.

Try related models on Kie.ai: Gemini 3.7 Flash, Gemini 3.8 Flash and Gemini 3.6 Flash.