AI Giants Slash Pricing as Enterprises Embrace Open-Weight Alternatives

Key Takeaways

• Enterprise adoption of open-weight AI models is accelerating as businesses seek alternatives to expensive proprietary systems from leading providers.
• Corporate discussion of open AI models surged six times higher during August and September versus the prior year, per AlphaSense analysis referenced by the Financial Times.
• Telecommunications giant AT&T currently processes approximately 40% of AI tasks using open models, with plans to expand this to 70% over the next twelve months.
• OpenAI introduced GPT-6 Sol and GPT-6 Luna models with API pricing reduced by 50% compared to previous offerings.
• Anthropic launched Claude Opus 5.5, featuring enhanced token efficiency with approximately 40% lower operational costs versus earlier versions.

A growing number of enterprises are shifting away from premium AI offerings to reduce expenditures. Rather than subscribing to high-end versions of services like ChatGPT or Claude, organizations are adopting open-weight models—AI frameworks that publish their learned parameters publicly, enabling users to modify and adapt them.

Operating open-weight systems typically requires lower investment than running cutting-edge proprietary models. This cost advantage is gaining traction as organizations prioritize budget management for their AI initiatives.

AlphaSense analytics reveal that executive discussions about open models during corporate presentations increased sixfold in August and September when compared with identical months in the previous year. These statistics were highlighted in Financial Times coverage.

Enterprise Leaders Detail Their Migration Strategy

Robert Reilly, serving as CFO at PNC Financial, revealed this month that his organization operates its own computing infrastructure. Reilly explained PNC leverages frontier models via leading cloud platforms while simultaneously maintaining open-weight systems within proprietary data centers.

Arun Rajan, chief strategy and innovation officer at C.H. Robinson, indicated the organization intends to progressively route less complex operations to open-source frameworks.

Spencer Rascoff, leading Match Group as CEO, noted his organization represents a significant client for major cloud service providers. Rascoff mentioned Match additionally employs Chinese open-weight systems as a cost management strategy.

Vinay Kuruvila, technology chief at Tinder, stated that proprietary frontier models currently satisfy 90% of his team’s requirements. Kuruvila suggested that continued advancement in open-weight alternatives could eliminate the need for frontier models entirely.

Andy Markus, chief data and AI officer at AT&T, disclosed the telecommunications company presently executes roughly 40% of AI operations through open models. Markus outlined AT&T’s objective to elevate this proportion to 70% over the coming year.

According to Markus, continuous enhancements in open model capabilities provide the organization with expanded flexibility for workload deployment decisions.

Pricing Isn’t the Sole Consideration

In addition to financial benefits, certain organizations emphasize that open models deliver superior control regarding data governance and security protocols. Data center provider Digital Realty developed an internal communication platform powered by open models.

Scott Wallace, senior global director at Digital Realty, explained the organization inputs confidential information into its private model that would never be shared with frontier systems. Wallace emphasized that client data remains completely isolated from frontier model environments.

This week, both Anthropic and OpenAI countered competitive pressure by launching more affordable options. OpenAI expanded its GPT-6 lineup with two additional tiers, designated Sol and Luna, reducing API costs by half relative to previous pricing structures.

Sol targets sophisticated applications including software development. Luna focuses on high-volume operations such as document synthesis or data extraction.

Anthropic introduced Claude Opus 5.5, positioning it as a more token-optimized solution. The provider stated operational costs will decrease by roughly 40% compared to the previous Opus iteration.

Dianne Penn, product management lead at Anthropic, informed CNBC that the organization prioritizes engineering models to generate responses using minimal tokens, adjustable based on user-selected effort parameters.

These product launches represent the first major releases from either laboratory since Anthropic CEO Dario Amodei advocated for industry-wide deceleration in advanced AI development. Both organizations now face the challenge of retaining clients amid intensifying competition from economical open-weight alternatives offered by companies including Alibaba, Moonshot AI, and DeepSeek.