● Covered by 1 source · 1 reportMedium impact1 neutral
Toast 1 Search Agent Released, Outperforms GPT-5.6 Sol and Claude Opus 5 in Cost and Speed
Key points
- Toast 1 matches or outperforms Claude Opus 5 and GPT-5.6 Sol in search quality.
- It is up to 10 times cheaper and 12 times faster than comparable models.
- Toast 1 achieved 70% answer correctness on OfficeQA Pro V2, setting a new state-of-the-art.
- It can function as a standalone agent or as a sub-agent for frontier models.
Introduction of Toast 1
Toast 1, the first specialized search agent, is now available. It delivers frontier search quality, performing comparably to or better than Claude Opus 5 and GPT-5.6 Sol. This new agent is designed to be more efficient, operating up to 10 times cheaper and 12 times faster than these established models. While optimized for Mixedbread Search, it is compatible with any search backend.
Addressing the Cost of Frontier Models
Frontier AI models are capable of complex knowledge work, including reasoning, analysis, and information retrieval from extensive document collections. However, these models represent the most expensive component in the AI stack. As the use of AI intelligence becomes more metered, there is an increasing demand for specialized agents that can match the capabilities of frontier models at a reduced cost.
Specialized Retrieval and Agentic Labor
Toast 1 functions by taking over the entire search loop. It decomposes initial queries into subqueries, gathers evidence, inspects sources, and curates relevant context before returning it. This specialization allows generalist, frontier-level models to focus their context and compute resources on reasoning, acting, and producing final answers, rather than on the search process itself. This division of labor leads to more cost-effective search operations and improved end-to-end results on various tasks.
Performance Benchmarks
On the OfficeQA Pro V2 benchmark, which evaluates answer correctness in enterprise financial situations, GPT-5.6 Sol with Toast 1 as a sub-agent achieved 70% correctness at approximately $1.15 per task. This score represents the highest among systems evaluated by Databricks, establishing new state-of-the-art performance in both quality and efficiency. In comparison, Claude Fable 5 on Databricks Genie previously achieved 60% correctness at about $4 per task, while GPT-5.6 Sol without Toast 1 only reached 33% correctness.
Economic Impact of Evidence Gathering
The performance improvement of Toast 1 is attributed to a reformulation of the economics of evidence gathering. Its specialization enables the production of high-quality, token-efficient evidence packages. This efficiency leaves ample resources for the reasoning process to effectively reach the final answer, optimizing the overall AI workflow.
✨ This summary was generated by AI from the outlets' reporting listed below. It is not independently verified and may contain errors — check the original sources. How BrevFeed works →
The daily brief
One email each morning: the day's tech stories, clustered across outlets and summarized. No account needed.
One email a day. Unsubscribe in one click, any time.
Today's brief
Spend a few minutes, get the whole day. Every topic's top stories in one hands-free rundown — listen, watch, or read the transcript.
▶ Play today's brief
New every morning, and the back catalogue is archived by date.
Primary sources
GitHub mixedbread-ai/toast-harnessReporting from
Toast 1, a specialized search agent, has been released, offering frontier search quality at a lower cost and faster speed than existing models like Claude Opus 5 and GPT-5.6 Sol. This development is significant because it provides a more economical way for AI models to perform knowledge work by specializing the search loop, freeing up generalist models for reasoning and final answer generation.