Google released Gemini 3.7 Flash on 13 August 2026, twenty-three days after Gemini 3.6 Flash. Google calls it “our most intelligent workhorse model yet for coding and agents.” The benchmark deltas over its three-week-old predecessor are large, and the introductory price is half what 3.6 Flash cost. The interesting fact here is not any single number; it is the cadence.

What happened

Gemini 3.7 Flash is the third Flash-tier release in the 3.x line this year, after Gemini 3.5 Flash at I/O on 19 May 2026 and Gemini 3.6 Flash on 21 July 2026. Google positions it as the default workhorse: not the frontier reasoning tier, but the model most production traffic is expected to run on.

Google’s published comparison against Gemini 3.6 Flash:

BenchmarkGemini 3.7 FlashGemini 3.6 Flash
FrontierCode 1.143.6%34.4%
DeepSWE v1.165.3%49.0%
WebDev Arena (Elo)15881538
GDP.pdf34.0%22.0%
AutomationBench30.4%17.0%

These are Google’s own evaluations. The independent Artificial Analysis Intelligence Index placed Gemini 3.7 Flash at 56 — useful as a cross-check, because vendor-run agentic benchmarks are among the least comparable numbers in the industry.

Pricing. Introductory rates of $0.75 per million input tokens and $3.75 per million output tokens run through 31 December 2026, rising to $1.50/$7.50 on 1 January 2027. The standard rate is what 3.6 Flash charged, so the introductory period is a genuine 50% discount rather than a repricing — third-party listings show it flagged as exactly that. The model is documented with a 1,048,576-token context window and up to 65,536 output tokens.

Availability. Google Antigravity, the Gemini API through AI Studio and Android Studio, the Gemini Enterprise Agent Platform, and Gemini Spark for Pro and Ultra subscribers across 160+ countries.

The rest of Google’s August

Gemini 3.7 Flash did not ship alone. In the same month Google announced:

  • Gemini 3.5 Transcribe, a speech-to-text model aimed at voice agents, live captioning, and post-call analytics, with better handling of noise and technical jargon.
  • Gemini Omni 1.1 Flash, a video generation model adding scene extension, frame interpolation, and 4K upscaling, in Flow, AI Studio, and the Gemini app.
  • The Gemini app passing one billion monthly active users, which Google describes as the fastest-growing product in its history.
  • Gemma passing one billion cumulative downloads.
  • WeatherNext 2, an open-sourced cyclone prediction model.
  • The Pixel 11 series on Tensor G6, running Gemini Nano on device.

Why it matters for builders

Three weeks between workhorse releases changes how you should hold model choice.

Pin your model version. A 16-point jump on DeepSWE is welcome; an unannounced behaviour change under an unpinned alias is not. If your prompts, evals, or tool-calling logic were tuned against 3.6 Flash, treat 3.7 as a migration with its own eval run, not a free upgrade.

Budget on the standard price, not the introductory one. $0.75/$3.75 expires on 31 December 2026. A cost model built on introductory rates doubles in January. This is the second time in a month a lab has moved a headline price — Anthropic went the other way and made Sonnet 5’s introductory pricing permanent — which is a reminder that inference pricing is currently a competitive instrument, not a stable input.

The Flash tier is where the interesting competition is. The frontier tier gets the headlines, but a model at $0.75 per million input tokens scoring 65.3% on DeepSWE is what changes unit economics for agent loops, where every step pays for the last step’s output. See model tier routing for how to exploit that without hardcoding a single vendor, and LLM cost optimization for the arithmetic.

Sources

  1. Google, “Gemini 3.7 Flash: our most intelligent workhorse model” (13 August 2026): https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-3-7-flash/
  2. Google, “The latest AI news we announced in August 2026”: https://blog.google/innovation-and-ai/technology/google-ai-updates-august-2026/
  3. Google Cloud, “Gemini 3.7 Flash” model documentation: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/3-7-flash
  4. Google, Gemini API release notes: https://ai.google.dev/gemini-api/docs/changelog
  5. OpenRouter, Gemini 3.7 Flash model card (context length, output limit, live pricing): https://openrouter.ai/google/gemini-3.7-flash
  6. 9to5Google, “Gemini 3.7 Flash launches three weeks after last model, live in Spark” (13 August 2026): https://9to5google.com/2026/08/13/gemini-3-7-flash-launch/
  7. TechCrunch, “Google releases three new Gemini models — but no 3.5 Pro” (21 July 2026): https://techcrunch.com/2026/07/21/google-releases-three-new-gemini-models-but-no-3-5-pro/

Further reading