Sakana: Sakana Namazu
sakana's efficiency model. Context window: 262K tokens.
Scores by test
Methodology →What you need to know
Sakana Namazu is positioned as a high-cost model relative to its performance metrics. With a blended cost of $3.24 per million tokens, it operates at a price point that typically correlates with high-tier reasoning models, yet it currently ranks last among all 164 models evaluated in the dataset.
The model exhibits significant instability in its safety calibration, scoring a 1/5. This indicates a failure to distinguish between harmful and legitimate requests, which typically manifests as the model over-refusing benign prompts. This creates a friction-heavy user experience where the model may decline to answer standard queries.
Despite a generous 262K context window, the model's overall internal score of 1.00/5.0 suggests it cannot effectively leverage this capacity for high-quality output. Given the combination of poor performance and premium pricing, it offers low value per token.
Use this model if you require a specific, niche architectural implementation provided by Sakana. Skip this model if you need reliable output, cost-efficiency, or a model that does not over-refuse benign prompts.
Strengths — Top 3
Relative weaknesses — Bottom 3
Similar models