Stealth: Claude Sonnet 4.6 (20% off)

This offering is positioned as a reasoning-oriented model for agent-style workflows that require chain-of-thought planning and tool use. The listing is framed around cost-effective multi-step reasoning, native tool calling support, and adjustable temperature control, making it suitable for workflows where an application needs to orchestrate external calls while maintaining conversational context. The closed-weights flag indicates the underlying model is not openly distributed; it is served through a managed API rather than being deployable locally.

The aggregator description emphasizes a very large context window suitable for processing long documents or extended codebases in a single session, paired with a sizeable output budget for generating sustained responses. Cached read pricing is notably lower than standard input pricing, which benefits repeated queries against the same prompt material, and cache write pricing sits between input and read rates. The "20% off" branding and "Stealth" prefix are Kilo Gateway-specific labels, so the discount applies to token rates on that gateway rather than reflecting a separate upstream model version.