Google: Gemma 4 31B
Capabilities
Strengths
- Exceptional vision capabilities with video understanding
- Strong reasoning and long-document analysis (256K context)
- Open weights enable self-hosting and full data control
- Excellent instruction-following and information extraction
- Multimodal strength: text, image, and video in one model
Weaknesses
- Code generation noticeably weaker than text/reasoning tasks
- Limited transparency on training data and methodology
- Slower inference than smaller budget models
- Output pricing ($0.34/M) higher than comparable open models
Pricing
Input / 1M tokens
$0.09
Output / 1M tokens
$0.33999999999999997
Context window
262.144k
Transparency
Open weights
10.0 / 10
Open training data
0.0 / 10
Open methodology
7.5 / 10
Licence openness
7.0 / 10
Provider disclosure
6.5 / 10
FMTI company score
N/A
Open-weight model from Google DeepMind with published weights, but limited disclosure of training data and methodology. Google provides some technical documentation but less comprehensive than research papers. Can be self-hosted for full privacy control.
Sustainability
Inference energy
N/A
Training footprint
N/A
Provider infrastructure
7.5 / 10
Google operates significant renewable energy infrastructure (70%+ renewable electricity), supporting inference efficiency. 31B parameter scale is moderate, reducing inference energy relative to larger models. Training footprint not publicly disclosed.