Introducing Toast 1
Points and comments are a snapshot, not live.
Toast 1 is a specialized search agent matching frontier model quality at lower cost and latency.
Mixedbread released Toast 1, a specialized search agent that matches or outperforms Claude Opus 5 and GPT-5.6 Sol on retrieval tasks while being up to 10x cheaper and 12x faster. It decomposes queries into subqueries, gathers evidence, inspects sources, and curates context. In financial benchmarks (OfficeQA Pro V2), GPT-5.6 Sol with Toast 1 reached 70% accuracy at ~$1.15 per task, besting prior SOTA. In legal benchmarks, it cut token usage by 3.5x while maintaining identical answer quality. Toast 1 costs $0.016-$0.023 per query with 8-second median latency, works with any search backend, and is available via Mixedbread API.
What commenters are saying
Commenters were skeptical of the branding, with several noting the name Toast sounds like a joke or a stroke reference. Others questioned how this differs from existing RAG pipelines. Mixedbread's founder clarified that Toast 1 is a specialized model co-designed with their indexing layer, and that it can plug into existing stacks. A few commenters praised the concept of specialized search agents but lamented Google's decline. Some expressed nostalgia for older search methods, while others asked practical questions about on-premise deployment and integration.