Gemini API CLI (Google)

What Gemini API CLI (Google) does

Google's official command-line tool ('gemini-api') for calling the Gemini API from a terminal or from any coding agent with shell access (Claude Code, Codex, Cursor, Antigravity and others). It isn't a coding agent: it's a stateless, zero-dependency binary that generates images, video, speech and music to files on disk, analyses PDFs, images, audio, video and YouTube URLs, transcribes recordings to md/json/SRT, and runs Managed Agents such as Deep Research in the background. It's built for agents, with --help/--usage, --schema, an offline --dry-run and JSON output.

Good for

• Install it (curl -fsSL raw.githubusercontent.com | bash), set GEMINI_API_KEY, then ask Claude Code: 'use the gemini-api CLI to transcribe meeting.mp3 to SRT' (gemini-api transcribe -i meeting.mp3 --format srt --out meeting.srt)
• Summarise a long PDF contract or report from the terminal: gemini-api analyze -i report.pdf 'list the key risks'
• Let your coding agent create a hero image and a voiceover for a landing page while it builds the page (images need billing on)
• Kick off a background Deep Research run on a competitor or market and collect the JSON report later (roughly $1-3 per run)
• Run gemini-api generate 'hello' --dry-run to see exactly what a call will send before you spend anything

Honest take

Some claims not independently checked

• Repo google-gemini/gemini-api-cli, Apache-2.0, ~324 stars, labelled 'experimental'.
• Blog: aistudio.google.com .
• Install: curl install script (macOS/Linux), PowerShell script (Windows), or go install.
• There is no npm or pip package.
• The pitch 'one CLI gives any agent Gemini's media and research powers' holds up, but the headline media features (images, video, music) need a paid (billing-enabled) key, Deep Research costs real money per run, and the command surface may still change.
• UK/EU: the Gemini API is available in the UK and EU.
• Google's terms let you use free quota for your own development, with paid-tier data handling applied in EEA/CH/UK.
• If you offer an app to users in the EEA, Switzerland or UK, it must use Paid Services.
• Dev-level tool, so non-coders will reach it through a coding agent.

What we couldn’t verify

• Fact vs hype: Real and official.

Our Hype or Fact? checks

• FactGemini API CLI: Google's media models in one terminal command 27 Oct

Pricing

The CLI is free and open source (Apache 2.0). You pay for Gemini API usage on your own API key. Free tier: text models such as Gemini 3.8 Flash and Gemini 3.8 Flash TTS are 'free of charge' with quotas. Paid tier only: Nano Banana 2.1 images (~$0.034 per 1K image), Omni 1.1 Flash video (~$0.10/second of 720p), Lyria 3.5 music ($0.08/song). Deep Research is billed at standard token and tool rates, which Google estimates at ~$1-3 per task (Max ~$3-7); no free tier is stated. Gemini 3.8 Flash paid rates double on 1 Jan 2027.

Prices as listed by the maker when we last checked (2026-10-27); usually in US dollars. Check the website for current prices.

Latest update

• 2026-10-26: Experimental preview (v0.4.x) launched on X by Google DeepMind's Philipp Schmid (@_philschmid). The AI Studio blog post is dated 19 Oct 2026. @hqmank called it a smart move that lets Claude Code/Codex keep coding while Gemini handles media, analysis and research.
• 27 October 2026: New entry: Google's experimental Gemini API CLI (gemini-api), an open-source (Apache-2.0) zero-dependency binary that gives Claude Code, Codex, Cursor and other agents Gemini media generation, PDF/YouTube analysis, SRT transcription and background Deep Research. Images, video and music need paid billing. Commentary by @hqmank.

Is Gemini API CLI (Google) worth it? Quick answers