en.Wedoany.com Reported - FAR Labs has announced the opening of registration for its FAR AI inference platform, granting developers access and publishing pricing for multiple model deployments. The Abu Dhabi-based AI infrastructure company (a subsidiary of Dizzaract) aims to help developers reduce the operational costs of AI applications as usage of software products and automation tools grows.
At the core of the platform is a distributed inference network that matches builders' needs with available computing resources. Users can access the system via an OpenAI-compatible API, select from multiple models, and launch quickly, with workloads routed by the FAR Orchestrator to GPU resources for execution. FAR Labs positions cost as its primary competitive advantage, claiming through benchmark comparisons with other providers' pricing that its network structure enables significantly lower prices on certain deployments.
Specifically, for the Qwen3-30B-A3B model, FAR AI lists a price of $0.03 per million tokens, compared to $0.35 for NextBit and $0.27 for DeepInfra. The company claims this pricing is up to 91% lower than the latter. For the Qwen2.5-72B-Instruct model, the company lists an FP8 price of $0.17 per million tokens, compared to $0.39 for NovitaAI BF16 and $0.38 for DeepInfra FP8. This reportedly makes its pricing 55% to 56% lower than the listed alternatives. For the Qwen3.5-122B-A10B model, FAR AI lists an FP8 price of $0.51 per million tokens. According to the company, output token costs are reduced by up to 79% compared to providers such as AtlasCloud FP8 and SiliconFlow FP8.
FAR Labs' move comes as AI companies face different economic challenges than in earlier years. Although the unit price per token has dropped significantly, total inference spending continues to rise as enterprises push more AI requests through customer support tools, agents, assistants, games, and internal workflows. This shift is particularly important for developers relying on proprietary APIs from companies like OpenAI and Anthropic. As products scale, recurring inference costs can compress margins and limit room for testing and expansion.
FAR Labs states that its pricing advantage comes from leveraging underutilized computing resources rather than relying on large dedicated data center clusters. The platform uses available GPU capacity from consumer devices and small-to-medium enterprise data centers, distributing workloads through its performance-focused orchestration layer. The company also aims to address issues beyond price, stating that builders running production workloads need systems that are private, reliably routed, low-latency, and practical for real-world scenarios. Its orchestration layer includes secure inference based on trusted execution environments, reliability scoring, support for open-source and proprietary models, and semantic vector flows. According to FAR Labs, the routing system is designed around uptime, workload continuity, and latency-sensitive performance.
The company says that during the SuperAI Singapore conference, multiple conversations with developers, GPU suppliers, model teams, investors, and corporate executives referenced interest in this type of infrastructure, pointing to a demand for faster, more reliable, lower-cost, and production-ready inference infrastructure. FAR AI is currently in a testing phase with closed partners and offers early access registration for builders, providing 1 million free tokens to those who join the program.
In a statement, Dizzaract founder and CEO Ilman Shazhaev outlined the company's market rationale. "The price of AI keeps dropping. Since 2021, the cost per token has fallen by about 99%. Yet AI bills keep rising because usage is growing faster than prices are falling. Inference is becoming the single largest cost in AI," he said. "Our cost advantage is not a discount from burning cash. It is structural."
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
