Gemini 3.7 Flash on RodiumAi: Google's workhorse for coding, agents, and video input, at half price

Google's Gemini 3.7 Flash is live on RodiumAi as google/gemini-3.7-flash. It is a 1M-context multimodal model for coding and agents, it accepts video as input, and intro pricing is 50% off through 31 December 2026.

What is Gemini 3.7 Flash?

On August 13, 2026, Google introduced Gemini 3.7 Flash: its most intelligent Flash workhorse yet for coding and agents. The release lands three weeks after Gemini 3.6 Flash. Google describes 3.7 as an algorithmic upgrade, not a new architecture: more discipline on multi-step plans, better tool use, fewer retries.

It is already live on RodiumAi as google/gemini-3.7-flash.

Flash is Google's production default: fast, cheap enough to run agents all day, strong enough for real software work. 3.7 Flash is the version you want if your stack mixes code, documents, screenshots, audio, and video, and you still need text in / text out with thinking.

What 3.7 Flash is built for

Coding and long-horizon agents

Google's launch numbers vs 3.6 Flash (Google blog):

Benchmark3.6 Flash3.7 FlashFrontierCode 1.1 Main (production-ready code)34.4%43.6%DeepSWE v1.1 (software engineering)49.0%65.3%WebDev Arena (Elo)15381588GDP.pdf (dense document comprehension)22.0%34.0%AutomationBench (enterprise workflows)17.0%30.4%

The pattern is consistent: first-pass code quality, debugging, issue resolution, and agent loops that actually finish. Google also reports better instruction following and more careful multi-step planning, which matters more than a single chat reply when you run tools in a loop.

Video, audio, and multimodal input

This is the practical upgrade many teams will feel first. On RodiumAi, Gemini 3.7 Flash accepts:

text
image
document (PDF and similar)
audio
video

Output is text. It is a video-in model, not a video generator. Use it to watch a product demo, a screen recording, a lecture clip, or a field video, then return a transcript-quality summary, a bug list, a QA checklist, or structured JSON.

For video generation, look at Gemini Omni Flash or Veo on the same catalog. 3.7 Flash is the model that reads video and reasons over it.

1M context and thinking levels

Context: 1,048,576 tokens
Max output: 65,536 tokens (about 66K on the model page)
Thinking: low, medium, high (minimal is not supported on 3.7)

Thinking tokens are billed as output. For cheap classification, keep thinking low. For autonomous tool loops, raise it.

Developer-facing details are in Google's 3.7 Flash developer guide.

50% intro pricing (through 31 Dec 2026)

Google is running introductory Standard pricing through 31 December 2026: half the original 3.6 Flash list rate.

Through 31 Dec 2026From 1 Jan 2027Input (per 1M tokens)$0.75$1.50Output (incl. thinking)$3.75$7.50Cached input$0.075$0.15

On RodiumAi, live RODI rates on the Gemini 3.7 Flash page currently track that wholesale intro:

RateRODI / 1MUSD (ref.)In479.4~ $0.75Out2396.9~ $3.75Cached48.0~ $0.075

RODI includes RodiumAi markup and upstream fees. USD figures are wholesale reference rates. After 31 December 2026, expect the catalog to move toward the $1.50 / $7.50 list unless Google extends the promo.

For agent traffic, this window is the cheap time to migrate from 3.6, scale evals, and lock prompts before the January step-up.

Gemini 3.7 Flash on RodiumAi

RodiumAi lists the model as google/gemini-3.7-flash.

On the model page you will find:

RODI pricing (input, output, cached) with USD reference rates
Capabilities: streaming, tool calling, vision, JSON mode, reasoning
Context: 1.0M · Max output: 66K
Inputs: text, image, document, audio, video
Upstreams: Vertex (default) and Google AI Studio (fallback)

Why use it through RodiumAi?

One API key for Gemini 3.7 Flash alongside GPT, Claude, DeepSeek, Mistral, Veo, and the rest of the catalog.
RODI billing: recharge via Mobile Money (Orange, MTN, Wave…) or bank transfer, without a Google Cloud invoice for every experiment.
OpenAI-compatible API at https://api.rodiumai.io/v1: keep your existing SDK or agent stack.
Failover: Vertex first, Gemini API second, same public slug.
Usage visibility in the dashboard: per-request cost, model breakdown, API key scoping.

Quick API example

from openai import OpenAI

client = OpenAI(
    base_url="https://api.rodiumai.io/v1",
    api_key="rd_sk_prod_…",
)

response = client.chat.completions.create(
    model="google/gemini-3.7-flash",
    messages=[
        {
            "role": "user",
            "content": "Plan a 4-step agent that reviews a PR, runs tests, and opens a follow-up issue if CI fails.",
        }
    ],
    max_tokens=4096,
)

print(response.choices[0].message.content)

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.rodiumai.io/v1",
  apiKey: process.env.RODIUM_API_KEY,
});

const res = await client.chat.completions.create({
  model: "google/gemini-3.7-flash",
  messages: [
    {
      role: "user",
      content: "Turn this meeting recording into action items with owners and deadlines.",
    },
  ],
});

console.log(res.choices[0].message.content);

Video in, text out

Pass a short clip as a data URL (mp4, webm, etc.). The model returns text: summary, bugs, chapters, JSON.

import base64
from pathlib import Path
from openai import OpenAI

client = OpenAI(
    base_url="https://api.rodiumai.io/v1",
    api_key="rd_sk_prod_…",
)

raw = Path("demo.mp4").read_bytes()
video_url = "data:video/mp4;base64," + base64.b64encode(raw).decode("ascii")

response = client.chat.completions.create(
    model="google/gemini-3.7-flash",
    messages=[
        {
            "role": "user",
            "content": [
                {
                    "type": "text",
                    "text": "Watch this product demo. List UI bugs, missing copy, and three follow-up test cases.",
                },
                {"type": "image_url", "image_url": {"url": video_url}},
            ],
        }
    ],
    max_tokens=2048,
)

print(response.choices[0].message.content)

Tip: Keep clips short for cost and latency. Video tokens count as input. Thinking tokens count as output. For repeated system prompts, cached input ($0.075 / 1M during the intro window) is the cheap path.

When to choose 3.7 Flash vs other models on RodiumAi

Use caseSuggested directionCoding agents, PR review, tool loopsGemini 3.7 FlashScreen recordings, lectures, field video → textGemini 3.7 FlashAlready on 3.6 with locked evalsStay on 3.6 until you re-bench, then moveMaximum Flash intelligence after the intro windowStill 3.7, budget for $1.50 / $7.50 from Jan 2027Cheaper high-volume extractionGemini 3.5 Flash-Lite / 3.1 Flash-LiteGenerate video (not just read it)Gemini Omni Flash or VeoLong-horizon premium codingClaude Fable 5 / Opus 5

3.7 Flash is a default production Flash, not a rare frontier call. Use it where agents, multimodal input, and the 2026 promo line up. Do not route every toy prompt through high thinking.

The bigger picture

Gemini 3.7 Flash sits on three 2026 trends:

Flash as the agent runtime: the cheap model is now the one that ships production code and runs tools.
Video as a first-class input: product QA, education, and field ops no longer need a separate transcription pipeline before the LLM.
Time-boxed intro pricing: Google is buying adoption through 31 December 2026, then stepping back to the standard Flash list.

For builders in Africa and beyond, RodiumAi keeps the same slug, local payment rails, and unified RODI billing.

Explore live pricing on the Gemini 3.7 Flash model page, create a key in your dashboard, and run one real agent or one real video clip before you swap 3.6 in production.