Gemini 3.7 Flash API: Pricing, Benchmarks, and How to Access It

Gemini 3.7 Flash is Google's new default fast model, released August 13, 2026, just three weeks after Gemini 3.6 Flash. The pitch is unusually direct: clearly better at coding and agent work than its predecessor, at half the launch rate. It streams around 340 output tokens per second, takes text, image, audio, video, and PDF input into a 1M-token context, and is callable today on Unifically as google/gemini-3.7-flash.

TL;DR: Gemini 3.7 Flash scores 43.6% on FrontierCode 1.1 (3.6 Flash: 34.4%), 65.3% on DeepSWE v1.1 (from 48.6%), and 97.0% on 128k long-context recall, while streaming ~340 tokens per second. Google launched it at an introductory rate, half what 3.6 Flash cost, holding through December 31, 2026. Call it as google/gemini-3.7-flash on the OpenAI-compatible /v1/chat/completions endpoint; billing is pay-per-use with live rates on the pricing page.

Key takeaways

• Released August 13, 2026, three weeks after Gemini 3.6 Flash, and positioned as Google's coding and agent workhorse rather than a chat model.
• The coding jump is the story: FrontierCode 1.1 rises from 34.4% to 43.6%, DeepSWE v1.1 from 48.6% to 65.3%, and WebDev Arena from 1538 to 1588 Elo.
• Long context is the quiet strength: 97.0% recall on GDM-MRCR v2 at 128k tokens, ahead of GPT-5.6 Terra at 93.5% and Claude Sonnet 5 at 81.5%.
• Speed rises with quality: about 340 output tokens per second at every thinking level, against roughly 234 for 3.6 Flash.
• Google's launch discount holds through December 31, 2026, with the list rate doubling back on January 1, 2027. Most headline numbers come from Google's own eval run; the independent picture is thinner.

What is Gemini 3.7 Flash?

Gemini 3.7 Flash is the newest model in the Gemini line overall; as of August 2026 the Pro line still sits at Gemini 3.1, and Google has said the next Pro release is delayed. Flash is where Google's velocity currently lives. The model takes text, image, audio, video, and PDF input, returns up to 64k tokens of text, and offers three thinking levels (low, medium, high) with medium as default. The knowledge cutoff reaches March 2026 for code and technical domains.

Artificial Analysis scores it 56 on the Intelligence Index at high thinking, one point behind GPT-5.6 Terra Max, while running several times faster. Even at medium thinking it scores 53, one point above 3.6 Flash at its best.

Gemini 3.7 Flash benchmarks

The launch table is about agents and code, and the gains over a three-week-old predecessor are unusually large:

Against the field, FrontierCode 43.6% edges out Claude Sonnet 5 at 42.7% and GPT-5.6 Terra at 41.3%, a Flash-class model beating balanced-tier frontier models on a hard coding eval.

Now the losses, because they define where this model stops. GPT-5.6 Terra stays ahead on DeepSWE v1.1 (69.6% against 65.3%) and Terminal-Bench 2.1 (87.4% against 85.8%) in the same table. Most headline numbers (FrontierCode, DeepSWE, MRCR) come from Google's own evaluation run, and the independent picture is thinner: its Arena text score sits at 1490 Elo, rank #9, marked preliminary. One more operational change: the minimal thinking level from earlier Flash models is gone, so even the cheapest classification calls now carry some thinking cost.

Gemini 3.7 Flash vs Gemini 3.6 Flash

The upgrade is real across every row Google published, and speed goes up rather than down. Unless a pipeline depends on regression-tested 3.6 Flash behavior, new work belongs on 3.7. The Gemini 3.6 Flash page stays live for teams mid-migration; its 77.45% on Vals AI's independent SWE-bench Verified re-run is also a reminder that the 3.6 generation was already strong, which makes the three-week turnaround to 3.7 the notable part.

Gemini 3.7 Flash pricing

Google launched 3.7 Flash at an introductory rate: half what 3.6 Flash cost per million tokens, holding through December 31, 2026. From January 1, 2027 the list rate doubles back to standard. That end date is worth putting in a calendar, because any cost model built on the launch rate quietly breaks in the new year.

On Unifically, billing is pay-per-use per token with no subscription and no credit expiry; the live rate sits on the pricing page alongside every other model on the platform, and one key covers Gemini 3.7 Flash next to image, video, and audio models. New accounts get $0.20 of free balance to test with. One billing detail specific to this model: thinking tokens bill as output tokens even though only summaries are returned, so higher thinking levels raise real output volume.

How to access the Gemini 3.7 Flash API

Google surfaces it through the Gemini API, AI Studio, Android Studio, and Antigravity. On Unifically it runs as google/gemini-3.7-flash on the OpenAI-compatible chat completions endpoint:

curl -X POST https://api.unifically.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "google/gemini-3.7-flash",
    "messages": [
      { "role": "user", "content": "Summarize the attached contract and list every obligation with a deadline." }
    ]
  }'

Set "stream": true for streaming; function calling and structured outputs work through the standard request shape. Full parameter docs: docs.unifically.com/models/llm/google/gemini-3.7-flash.

Frequently asked questions

What model ID does Gemini 3.7 Flash use?

google/gemini-3.7-flash, called through the OpenAI-compatible POST /v1/chat/completions endpoint with a Unifically API key.

Is Gemini 3.7 Flash better than Gemini 3.6 Flash?

Yes, across Google's whole evaluation suite: FrontierCode 1.1 rises from 34.4% to 43.6%, DeepSWE v1.1 from 48.6% to 65.3%, 128k recall from 91.8% to 97.0%, and measured output speed from about 234 to 340 tokens per second.

How fast is Gemini 3.7 Flash?

Around 340 output tokens per second at every thinking level, per Artificial Analysis, with time to first token under a second at low thinking. That makes it one of the fastest frontier-adjacent models available.

What is the Gemini 3.7 Flash context window?

1M tokens of input context with up to 64k output tokens, and the best long-context recall in the Flash line: 97.0% on GDM-MRCR v2 at 128k and 62.5% on the 1M pointwise variant.

Is there a Gemini 3.7 Pro?

No. As of August 2026 the Pro line is still at Gemini 3.1 and Google has said the next Pro release is delayed. Gemini 3.7 Flash is the newest Gemini model overall.

How long does the launch pricing last?

Google's introductory rate holds through December 31, 2026; from January 1, 2027 the list price doubles back to standard. Unifically's live per-token rate is always on the pricing page.

What to watch next

Three dates and boards to track: December 31, 2026, when Google's launch discount ends and every cost comparison in this space needs rewriting; Arena's text board, where the preliminary 1490 Elo will settle as votes accumulate; and the delayed Gemini Pro line, since a 4.0-generation Pro would reset the top of Google's lineup. We will update this post as each lands.

If you are choosing between the fast tier and a frontier agent model, Grok 4.6 and GPT-5.6 Sol are the two current reference points, both live on the same API.