Mixedbread launches Toast 1, a deep-search agent it says undercuts frontier models on cost

Mixedbread launched Toast 1 on Aug 13, 2026, a purpose-built deep-search agent the company says matches or outperforms Claude Opus 5 and GPT-5.6 Sol on search benchmarks while costing far less to run. It runs up to 12 times faster and 7 to 11 times cheaper than those models, Mixedbread says, answering in a median of roughly 8 to 11 seconds.

The pitch is a bet against frontier generalists: that a task-specific agent, tuned only for retrieval, can beat far larger models on search work at a fraction of the price. Toast 1 breaks a question into subqueries, retrieves candidate evidence and verifies it before answering, and can run on its own or as a subagent inside a larger system.

Mixedbread’s own numbers carry the argument. The company says pairing Toast 1 with GPT-5.6 Sol reaches 70 percent answer correctness on a benchmark it calls OfficeQA Pro V2 at about $1.15 per task, and that Toast 1 uses 3.5 times fewer tokens than a baseline on a “Harvey LAB Firm Knowledge” test while matching its performance. The agent works best paired with Mixedbread Search, the company said, but is compatible with other retrieval backends.

Toast 1 is available immediately through the Mixedbread API, priced at $0.30 per million input tokens, $0.036 per million cached input tokens and $0.72 per million output tokens.

Every performance and cost figure here comes from Mixedbread and has not been independently verified. The company has not published the benchmark datasets in full, and OfficeQA Pro V2 is a Mixedbread-defined test rather than an external standard, so cross-model comparisons should be read as vendor claims.