What happened
AT&T announced that about 40% of its internal AI workloads are now routed through open‑ models such as Nvidia Nemotron, Meta Llama, and Google Gemma, using a custom routing system built on LiteLLM. The shift has reduced AI coding costs by 56% with only a 2% dip in output quality, and for some complex tasks the savings reach 80‑90% versus closed‑model alternatives. AT&T also launched OTel 2.0, a post‑trained open‑weight model for telecom data, developed with the GSMA’s Open Telco AI initiative.
According to Crypto Briefing, AT&T has implemented an intelligent routing system that directs AI requests to the most appropriate model based on task complexity. Simpler queries are sent to open‑ models—including Nvidia’s Nemotron, Meta’s Llama, and Google’s Gemma—while more demanding workloads continue to use closed‑source models when higher quality is required.
The company reports that this routing has cut AI coding costs by 56% with only a 2% reduction in output quality. For certain complex workloads, cost reductions are even higher, ranging from 80% to 90% compared with traditional closed‑model solutions.
AT&T’s AI consumption has surged from roughly 8 billion tokens per day a year ago to about 45 billion tokens per day now, a 5.6‑fold increase. To manage this growth, AT&T introduced a custom AI gateway built on LiteLLM, which matches each request to the most cost‑effective model.
In parallel with consumption, AT&T launched OTel 2.0, an open‑ model trained on more than 400 billion tokens of telecom‑specific data. The model was co‑developed with the GSMA’s Open Telco AI initiative and built using AMD GPUs and Microsoft’s Foundry platform.
Source details: cryptobriefing.com ↗
Why it matters
The move illustrates a growing enterprise trend toward open‑ AI to curb soaring AI expenses while preserving performance. By routing the majority of routine requests to cheaper, open models, AT&T demonstrates that large‑scale organizations can achieve substantial cost efficiencies without sacrificing quality, potentially reshaping procurement strategies for other telecoms and data‑intensive firms. The launch of a bespoke open‑weight model (OTel 2.0) signals that companies are not only consuming but also contributing to the open‑weight ecosystem, which could accelerate innovation and reduce reliance on proprietary providers such as Anthropic and OpenAI. However, the article does not disclose pricing details for OTel 2.0, nor does it provide independent verification of the reported quality metrics, leaving open questions about broader applicability and long‑term performance.
The cost savings reported by AT&T highlight the financial pressure enterprises face as AI usage scales. By demonstrating that open‑ models can handle a large share of workloads at a fraction of the cost, AT&T provides a practical blueprint for other large organizations seeking to manage AI spend.
AT&T’s decision to develop its own open‑ model (OTel 2.0) underscores a shift toward greater data sovereignty and customization. Companies can tailor models to industry‑specific data, reducing reliance on external vendors and potentially improving compliance with privacy regulations.
The reported 2% quality dip suggests that, at least for AT&T’s internal use cases, open‑ models are approaching parity with proprietary alternatives. If this performance gap continues to narrow, it could diminish the market advantage of closed‑source providers and stimulate more competition in the AI model space.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
In AI, what are a model's "parameters"?
What to watch next
Future updates on AT&T’s target of routing 60‑70% of AI workloads through open‑ models, the commercial availability and pricing of OTel 2.0, and any measurable impact on service quality or customer experience will be key indicators of the strategy’s success. Additionally, monitoring whether other telecom operators adopt similar routing architectures or develop their own open‑weight models will reveal whether this cost‑saving approach spreads across the industry.
Whether AT&T meets its internal target of routing 60‑70% of AI workloads through open‑ models within the next year, and how that transition impacts overall operational efficiency.
The commercial rollout plan for OTel 2.0, including pricing, licensing terms, and whether the model will be made available to external partners or remain an internal tool.
Adoption signals from other telecom operators or large enterprises that may emulate AT&T’s routing architecture or develop their own open‑ models, indicating broader industry movement.
Any measurable effects on service quality, customer experience, or regulatory compliance that can be directly linked to the shift toward open‑ AI.