Google DeepMind
London, United Kingdom · founded 2010 · Gemini models
Google's SI lab, formed by combining DeepMind with Google's Brain team. It builds the Gemini models and the open-weights Gemma family.
- Tracked models
- 5
- 1 with open weights
- Benchmarks led
- 3
- Vals Index, Arena Text, Finance Agent
- Latest release
- Sep 30
- Gemini 4 Argon
- Cheapest
- $0.15
- Gemma 4 31B, blended per 1M
Current lineup
5 models, strongest first.
Released
Google's newest and largest Gemini, built to keep reasoning through very long agentic jobs such as codebase migrations, financial and legal research, and finding and patching software vulnerabilities. It is not generally available yet.
- Index
- 52.6
- In / 1M
- $2
- Out / 1M
- $10
- Context
- 1M
Gemini 3.8 Flash
Mid-tierReleased
Google's current mainstream model: fast, low-cost, multimodal and built around a 1M-token context, aimed at coding agents and multi-step knowledge work. It ranks second on the Finance Agent benchmark despite sitting mid-table overall.
- Index
- 40.9
- In / 1M
- $0.75
- Out / 1M
- $3.75
- Context
- 1M
Released
The newest Pro-tier Gemini that developers can actually call: a heavier reasoning model for hard problem solving and multimodal analysis. It predates the current agentic benchmarks and now trails the newer Flash models on most of them.
- Index
- 29.7
- In / 1M
- $2
- Out / 1M
- $12
- Context
- 1M
Gemini 3.5 Flash-Lite
Small & fastReleased
Google's budget tier: a very fast, low-cost multimodal model for sub-agents, document processing and other high-throughput jobs. It is the fastest model tracked here by a wide margin.
- Index
- 22.2
- In / 1M
- $0.30
- Out / 1M
- $2.5
- Context
- 1M
Released
Google's strongest open-weights model: a 31-billion-parameter multimodal model with an optional thinking mode that fits on a single 80GB GPU, or on consumer GPUs when quantised. Permissively licensed and meant for self-hosting, fine-tuning and local agents.
- Index
- 14.7
- In / 1M
- $0.09
- Out / 1M
- $0.34
- Context
- 262K