Google DeepMind
Google's AI lab behind the Gemini model family, sold to developers through the Gemini API, Google AI Studio and Vertex AI
Overview
Google DeepMind is Google's AI research lab and the maker of the Gemini models. Developers reach those models through the Gemini API and Google AI Studio, or through Vertex AI on Google Cloud for enterprise controls. The current API lineup, per Google's model list, is Gemini 4 Carbon (described as "our strongest reasoning model, built for complex software engineering, multimodal knowledge work"), Gemini 4 Argon (an efficient frontier model), the Gemini 3.x Flash and Flash-Lite models, and Gemini 3.1 Pro in preview, plus image (Nano Banana), speech, embedding, music (Lyria) and robotics models. Gemini 4 takes text, images, audio, video and PDFs with a 1M-token context. Gemini 2.5 Flash and 2.5 Pro are deprecated and will be routed to newer models on December 1, 2026. DeepMind was founded in London in 2010 and merged with Google Brain in April 2023; Demis Hassabis is CEO. Also from Google DeepMind: Gemma, its open-weight model family; Google Gemini, the consumer assistant; Gemini Robotics, models for robot reasoning and control; and Genie, a real-time world model you can explore in Project Genie. See also: TensorFlow, JAX and Keras.
The Verdict
Who Should Use Google DeepMind's Gemini API?
Best For
- Apps that mix text, images, audio, video and long PDFs in one prompt
- Prototyping on a free tier before paying (Flash and Flash-Lite models)
- Teams already on Google Cloud who want Vertex AI governance
- High-volume, low-cost work on Flash-Lite models
Not Ideal For
- Teams that need stable model names for years (models are deprecated and rerouted often)
- Budgets that cannot absorb price changes (several prices rise on Dec 1, 2026 and Jan 1, 2027)
- Self-hosting (use Gemma for open weights)
What's Great
- Native multimodal input with a 1M-token context on Gemini 4
- Free tier on several Flash and Flash-Lite models
- Batch is about half the Standard price
- Users on Reddit rate multimodal and research work as a strength, and some find it cheaper than other frontier APIs
Watch Out For
- Gemini 4 prices double on December 1, 2026 ($4 in, $20 out per 1M tokens)
- Free-tier inputs may be used to improve Google products
- Developers still report that coding and agent work trail some rivals
- Frequent renames and shutdowns of preview models
Pricing
View all features & details
Current API Models
- Gemini 4 Carbon and Gemini 4 Argon (1M-token context)
- Gemini 3.8 Flash and 3.8 Live (new), 3.6 Flash (stable)
- Gemini 3.5 and 3.1 Flash-Lite (stable)
- Gemini 3.1 Pro preview: $2 in, $12 out per 1M up to 200K prompts
- Nano Banana image models, Lyria music, TTS, embeddings, Deep Research agents
Access and Billing
- Google AI Studio and the Gemini API for developers
- Vertex AI on Google Cloud for enterprise use
- Batch tier about 50% of Standard; Priority about 1.8x
- Grounding: 5,000 free Search or Maps requests a month on Gemini 3 and later, then $14 per 1,000
Company
- Founded 2010 in London by Demis Hassabis, Shane Legg and Mustafa Suleyman
- Acquired by Google, then merged with Google Brain in April 2023
- CEO: Demis Hassabis
Real-World Usage
Independent Benchmarks
- Artificial Analysis scores Gemini 3.1 Pro Preview at 46 on its Intelligence Index v4.1 (57 on the launch-era index)
- Scores depend on index version, so compare models on the same version
User Sentiment
- G2 reviews of Gemini average about 4.4 out of 5 (these cover the Gemini app, not the API)
- A common developer pattern: Gemini for research and documents, another model for coding
How It Compares
| Feature | Gemini 4 Carbon | Gemini 3.8 Flash | Gemini 3.1 Flash-Lite | Gemma 4 (open) |
|---|---|---|---|---|
| Role | Strongest reasoning | Fast general model | Lowest-cost API model | Self-hosted |
| Input price per 1M | $2 (to $4 on Dec 1, 2026) | $0.75 | $0.25 | Free weights |
| Output price per 1M | $10 (to $20 on Dec 1, 2026) | $3.75 | $1.50 | Free weights |
| Free API tier | No | Yes | Yes | Not needed |
| Context | 1M tokens | Not stated | Not stated | Up to 256K tokens |
Sources: Google DeepMind · Gemini models · Gemini API pricing · Vertex AI pricing · Artificial Analysis
Is this your tool? Add a badge