Gemini 3.7 Flash: Google's Most Capable Workhorse Yet

Gemini 3.7 Flash is now generally available at an introductory price of $0.75 per million input tokens and $3.75 per million output tokens — half of what the original Gemini 3.6 Flash cost. Google DeepMind introduced the model on August 13 as “our most intelligent workhorse model yet for coding and agents,” just three weeks after 3.6 Flash, and the Gemini API release notes confirm it as a GA release.

The model ships with the biggest single-release benchmark jump the Flash line has shown this year: FrontierCode 1.1 Main climbs from 34.4% on 3.6 Flash to 43.6%, and DeepSWE v1.1 from 49.0% to 65.3%. Google attributes the pace to “developer feedback and algorithmic innovations.”

Key facts

• GA on August 13, 2026 — gemini-3.7-flash released via the Gemini API; Google AI Studio, Android Studio, and Antigravity are the developer entry points.
• Introductory pricing: $0.75/1M input and $3.75/1M output tokens through December 31, 2026; $1.50/$7.50 from January 1, 2027 — half the original 3.6 Flash rate.
• Benchmarks vs Gemini 3.6 Flash: FrontierCode 1.1 Main 43.6% vs 34.4%; DeepSWE v1.1 65.3% vs 49.0%; WebDev Arena Elo 1588 vs 1538; GDP.pdf 34.0% vs 22.0%; AutomationBench 30.4% vs 17.0%.
• Updated safety safeguards for CBRN (chemical, biological, radiological, nuclear) and cyber-offense domains ship with the release.
• Consumer reach: Gemini’s Spark personal agent (Google AI Pro and Ultra subscribers) switches to 3.7 Flash, in over 160 countries.

What changed

Three weeks after 3.6 Flash, the 3.7 release closes the gap to Google’s reasoning-heavy tiers on agentic workloads without abandoning the Flash price band. The AutomationBench jump — 30.4 vs 17.0 — is the clearest signal: 3.7 Flash is being tuned for agents that plan, call tools, and retry, not just single-turn chat.

The pricing move is equally deliberate. Google is pricing 3.7 Flash at half the original 3.6 Flash rate through the end of the year, then stepping to $1.50/$7.50 in January. That is a timed discount, not a permanent cut — the same “introductory pricing” pattern OpenAI used with GPT-5.6 when it launched, and a signal that the price war that started in July is now fully symmetric.

Why it matters

• Agent economics just changed for developers. At $0.75/$3.75, a deep agent run that burns hundreds of thousands of tokens costs cents. Google is explicitly courting production agent builders, where Flash models already power the always-on agent tier.
• The workhorse tier is now a benchmark battleground. Gemini 3.7 Flash, DeepSeek V4-Pro, and DeepSeek V4-Flash all publish agent-benchmark numbers in the 60–90 range this month; the differentiated layer has moved from general intelligence to long-horizon task execution.
• Pricing discipline is spreading. Half-price introductory tiers, peak/off-peak billing, and $0.75-per-million workhorses mean the marginal cost of frontier-adjacent intelligence keeps falling — the trend we’ve tracked since the July price cuts.

What to watch

• The January 1 step-up. Whether the $1.50/$7.50 price holds — or gets extended like Anthropic’s $2/$10 rate — will be the clearest read on demand elasticity in the workhorse tier.
• Independent benchmark runs. DeepSWE 65.3 is vendor-reported; third-party harness runs will verify whether 3.7 Flash actually beats reasoning-tier models on agentic coding. Update: 3.8 Flash launched September 2 at the same $0.75/$3.75 intro price, claiming 3x long-horizon task completion.
• Spark rollout. 3.7 Flash behind Spark in 160+ countries puts the model in front of millions of consumers within days, which will surface real-world reliability data fast.

Official source

• Google DeepMind: Introducing Gemini 3.7 Flash (August 13, 2026)
• Gemini API: Release notes — Gemini 3.7 Flash GA

Updated August 16, 2026.