Introducing Toast 1

Toast 1, the first specialized search agent developed by Mixedbread AI, is now available. This innovative agent outperforms Claude Opus 5 and GPT-5.6 Sol in search quality while being up to 10 times cheaper and 12 times faster. It excels with Mixedbread Search, but can function with any search backend.

As large language models become more advanced, the demand for specialized agents that can match their capabilities at a lower cost is growing. Toast 1 can operate as a standalone retrieval agent or as one of many subagents for a frontier model. It manages the entire search process: breaking down queries, gathering evidence, verifying sources, and curating the most relevant context before delivering the final results.

This division of labor between specialized agents and general models results in both lower costs and improved end-to-end performance on various tasks. In a test called OfficeQA Pro V2, which evaluates answer accuracy in complex, realistic financial scenarios, Toast 1 with GPT-5.6 Sol as a subagent achieved 70% correctness for $1.15 per task, the highest score among evaluated systems.

In comparison, the previous best, Claude Fable 5 on Databricks Genie, reached only 60% correctness at approximately $4 per task, while GPT-5.6 Sol alone within Codex only hit 33% correctness. Toast 1's specialized evidence gathering allows it to produce high-quality, token-efficient search results, freeing up resources for the reasoning process.

In a test on legal knowledge benchmarking, where context-heavy legal research is crucial, Toast 1 preserved answer quality while reducing token usage by 3.5 times, leading to a 60% cost reduction. Toast 1 is a lightweight model, costing approximately $0.016-0.023 per query, with a median latency of 8 seconds. It can reach frontier model performance at a fraction of the cost and latency of other retrieval agents.

Toast 1 is available through the Mixedbread API at a discounted launch price, with backend agnostic compatibility. It can be integrated with existing retrieval indexes or used directly in agentic workflows through Chat Completions API, OpenCode integration, or npx skills add mixedbread-ai/skills. A free trial with $5 in included credits is available.