Gemma 4
Gemma 4
Google's open weights, from something that runs on a phone up to a real server model.
Updated
The read
Gemma 4 is the most useful open-weights family Google has shipped: Apache-licensed, multimodal, and released in sizes from a couple of billion parameters to around thirty, so the same family covers a laptop, a phone and a rack. Quality lands well above its parameter count, and the licence has none of the user-count clauses that complicate the alternatives. It is not going to beat a frontier API on a hard problem.
Where it shines
- Apache licence with no user-count or revenue clauses
- Sizes that genuinely fit a laptop, a phone and a server
- Multimodal at every size, which is rare in open weights
Worth knowing first
- Clearly behind frontier APIs on hard reasoning
- Google publishes no hosted first-party API for it at frontier scale
Index score
80.1
Provisional. This score is read from the product's public capability, not from the one-prompt rebuild the fully reviewed entries went through, so treat it as a placing rather than a verdict.
- Rank
- 25 of 53
- Builds
- LLMs
- Category
- Open weights
- Output
- Apache-licensed weights, several sizes
- Pricing
- Free tier
- Best for
- Running a capable model on hardware you control
Scores are our own editorial judgement, weighted craft 55, speed 10, control 15 and value 20.
Also worth a look
Full index- 05Kimi K3Kimi K3ProvisionalThe largest open-weights model anyone has shipped, and it reads a million tokens without blinking.Open weightsFree tier87.5
- 06GLM-5.2GLM-5.2ProvisionalFrontier-class output, an MIT licence, and a million tokens of context. That combination is the story.Open weightsFree tier87.1
- 12Qwen 3.6Qwen 3.6ProvisionalThe open Qwen release most people should actually run, in sizes that fit real hardware.Open weightsFree tier83.3
