-- Loxi, an Istanbul-based artificial intelligence company focused on professional knowledge work, today announced that its AI system achieved a 58.2% score on JobBench, placing it in the same performance range as several frontier AI models while operating at less than one-tenth of the cost of the largest systems on the benchmark. Loxi completed all 65 tasks in JobBench’s main evaluation split, producing a deliverable for every task without time-outs, refusals, or empty output folders.
The company said the result demonstrates how smaller, specialized AI models can compete with much larger general-purpose systems when the underlying model is paired with software designed specifically for professional workflows.
Loxi is currently available as a free AI product for knowledge-intensive tasks including spreadsheet analysis, financial research, legal memoranda, project planning, document creation, and other forms of office-based work.
Focus on the AI System Around the Model
Rather than relying primarily on larger models or additional computing resources, Loxi has focused its development on the software layer surrounding the model.
The company describes this layer as its AI harness, which determines how a task is planned, how documents and files are interpreted, which tools and skills are used, how intermediate work is checked, and how a final deliverable is produced.
Over the past eight months, the Loxi team has run thousands of controlled experiments examining the interaction between models and their surrounding agent systems.
Its research indicates that model selection alone does not determine performance. According to the company, matching a model with a task-specific harness can produce materially different results even when the underlying model, task set, and evaluator remain unchanged.
The approach reflects a broader shift in AI development toward improving agent architecture and domain-specific systems rather than relying exclusively on larger foundation models.
Recent research and industry experiments have similarly shown that changes to agent runtimes and task-specific software can materially affect benchmark performance without changing the underlying model.
Domain-Specific AI Gains Enterprise Attention
Loxi’s benchmark result comes as enterprises increasingly explore smaller and more specialized AI systems for professional applications.
Gartner has projected that more than half of enterprise generative AI deployments will use domain-specific models by 2027, compared with approximately 1% in 2024.
Companies in sectors such as finance and legal services have already begun experimenting with this approach. Bridgewater Associates has developed AI systems trained around its investment workflows, while legal AI companies including Harvey have explored task-specific model optimization for professional work.
Loxi is applying a similar philosophy to a broader category of knowledge work.
Its beta product continuously incorporates experimental versions of the company’s model and agent system, with user activity and feedback informing subsequent development.
Building AI From Istanbul
The company’s progress also comes amid increased investment in artificial intelligence infrastructure and talent in Türkiye.
The country’s National AI Action Plan for 2026 to 2030 includes targets for expanding data center capacity, attracting private-sector investment, increasing AI literacy, developing specialist talent, and supporting the development of sector-specific AI systems.
Loxi said its work illustrates how smaller engineering teams outside the world’s largest AI laboratories can compete on applied tasks by focusing on model efficiency, software architecture, and specialization.
Rather than attempting to build a general-purpose frontier model, the company is concentrating on tasks professionals perform repeatedly in real working environments.
Making Professional AI More Efficient
JobBench evaluates AI systems on practical professional tasks that require models to interpret instructions, work with files, reason across multiple steps, and produce usable outputs.
For Loxi, the benchmark is intended to measure something more closely aligned with real-world business use than conventional question-and-answer evaluations.
The company believes this distinction is increasingly important as businesses evaluate AI not only on model capability, but also on cost, reliability, and the quality of completed work.
Loxi’s current product remains free to use while the company continues testing new model and harness configurations.
The company said its longer-term goal is to demonstrate that professional-grade AI performance does not necessarily require the largest or most expensive model available.
About Loxi
Loxi is an Istanbul-based artificial intelligence company developing AI systems for professional knowledge work. Its technology combines specialized language models with task-specific agent architecture designed for workflows including spreadsheets, research, financial analysis, legal documents, project planning, and other professional tasks.
The company is currently operating Loxi as a free beta while continuing research into model efficiency, agent systems, and domain-specific AI.
Contact Info:
Name: Loxi
Email: Send Email
Organization: Loxi
Website: https://www.loxi.works/
Release ID: 89203620
Should any errors, concerns, or inconsistencies arise from the content provided in this press release that require attention or if a press release needs to be taken down, we kindly request that you immediately contact us at [email protected] (it is important to note that this email is the authorized channel for such matters, sending multiple emails to multiple addresses does not necessarily help expedite your request). Our efficient team will be at your disposal for timely assistance within 8 hours – taking necessary measures to rectify identified issues or providing guidance on the removal process. We prioritize delivering accurate and reliable information.
