Upstage: Solar Mini 4
Upstage
Capabilities
Strengths
- Exceptional 524K context window for long-document tasks
- Ultra-low inference cost ($0.20/M output tokens)
- Efficient 3B active parameters with MoE architecture
- Open weights enable self-hosting and full data control
Weaknesses
- Extremely slow inference speed (0.05 speed_score)
- Weak reasoning and analysis capabilities (0.39)
- Not suitable for latency-sensitive applications
- Limited capability depth vs. larger or flagship models
Pricing
Input / 1M tokens
$0.049999999999999996
Output / 1M tokens
$0.19999999999999998
Context window
524.288k
Transparency
Open weights
10.0 / 10
Open training data
0.0 / 10
Open methodology
6.5 / 10
Licence openness
7.0 / 10
Provider disclosure
5.5 / 10
FMTI company score
N/A
Open-weight model with publicly available weights. Upstage provides moderate documentation of training methodology and model architecture. Limited disclosure of full training data composition and energy footprint.
Sustainability
Inference energy
7.5 / 10
Training footprint
N/A
Provider infrastructure
5.5 / 10
Small active parameter count (3B) suggests low per-inference energy consumption relative to larger models. Training footprint and provider's overall energy sourcing not disclosed. Self-hostable open weights enable low-carbon private deployments.