Anthropic’s reported two-tier Claude rollout and rising AI subscription costs are fueling debate over whether the most powerful systems will remain broadly accessible or become expensive, restricted tools for select users.
By yourNEWS Media Newsroom
The cost of using the most advanced artificial intelligence systems is rising sharply, and critics say the shift could determine who has access to the strongest models as AI becomes more central to research, coding, business and public information.
Anthropic’s reported release of Claude Fable 5 for commercial users and Mythos 5 for restricted access has become a focal point in that debate, according to The Daily Bell. The new pricing cited in the report is $10 per million input tokens and $50 per million output tokens, roughly double the cost of Claude Opus 4.8.
The pricing change has drawn criticism from technology observers who describe it as a “rug pull,” arguing that users are paying more while potentially being shifted to older models when queries are considered high-risk. Anthropic reportedly says fewer than 5% of sessions trigger that kind of downgrade, but critics warn the share could increase over time.
The concern is not only cost. Critics say the structure signals a broader move toward controlled access, user verification and restrictions on what people may ask advanced models to process.
Anthropic’s reported system divides access between two model tiers. Fable 5 is described as the commercial model available through cloud API access, Amazon Bedrock, Google Cloud and Pro, Max and Team subscription tiers. Mythos 5 is described as a restricted model available to a smaller group of vetted users, including defense organizations and critical-infrastructure operators.
Industry critics argue the arrangement could create a separate AI environment for government-linked and high-security users while ordinary commercial users face more limits. The report said Mythos 5 is available to about 200,000 defense organizations and critical-infrastructure operators and claimed safety restrictions do not apply to that tier.
The public-facing concern is that advanced AI access may increasingly depend on identity, institutional status and acceptable use determinations. Some critics have described the approach as “shotgun KYC,” suggesting that companies may begin tying user identity and access level to the kinds of questions being asked.
The discussion has intensified because leading AI companies remain financially strained despite rapid growth. The report said OpenAI lost $5 billion in 2024, while Anthropic lost $5.3 billion. Even excluding training costs from OpenAI’s 2024 revenue, the company would still have lost $2.2 billion. Perplexity AI spent 164% of its revenue on cloud computing services in 2024.
Those losses help explain why providers are raising prices. AI coding tools that were once inexpensive are becoming significantly more costly. The report said GitHub Copilot previously charged $10 per month, while the cheapest usable tier of Claude Code now costs $100 per month. OpenAI has reportedly discussed charging $20,000 per month for PhD-level research agents with investors.
The economics are also affecting companies built on top of leading models. Cursor, a major AI coding application, reportedly sends 100% of its revenue to Anthropic and remains deeply unprofitable, underscoring how expensive it can be to provide advanced AI services when underlying model usage costs remain high.
The pricing trend has led some experts to predict that top-tier AI tools could cost more than $20,000 per user per year within two years. That would make the most powerful systems unaffordable for many independent researchers, small companies and ordinary users.
The dispute also reaches beyond pricing into questions of information access. Critics argue that AI systems trained on heavily filtered or officially preferred sources could narrow the range of perspectives users receive. The report compares the issue to earlier changes in internet search and content ranking, where “authoritative content” became more prominent and competing sources were pushed lower or excluded.
The concern is that large models may be trained to compress vast amounts of information into answers that appear coherent while filtering out contradictory data. Critics argue that could limit discussion on contested topics, including nutrition, herbal medicine, permaculture, health practices outside conventional medical recommendations, crime statistics, immigration data and recession indicators.
The report also connects AI restrictions to broader government pressure in Five Eyes nations, including Canada, where critics argue officials are moving against what they describe as “socially destabilizing content.” Some observers have asked whether Grok remains one of the last major platforms with fewer restrictions, though others remain skeptical that any large commercial model is fully uncensored.
Local AI models are emerging as the main alternative for users concerned about rising costs, identity controls or information restrictions. Smaller but capable multimodal language models with about 13 billion parameters can run on personal computers, and executable model files can be shared through torrents or decentralized networks such as BitTorrent.
Supporters of local AI say the approach lets users research, customize and preserve access without depending entirely on centralized platforms. As the cost of fine-tuning and training models falls, industry observers predict that building and distributing large language models could soon become far cheaper, potentially reaching about $20,000 within two years and declining further after that.
The report argues that future advances in understanding how the human brain performs learning reinforcement could further reduce the cost, energy consumption and time required to train neural networks.
The AI industry now faces two competing futures. One path leads toward expensive, highly controlled systems serving large companies, government users and vetted institutions. The other points toward decentralized tools that users can run locally, modify and share.
For now, the economics of generative AI are pushing major providers toward higher prices. Whether that results in a premium market for elite access or a new wave of open-source and local competition will shape not only AI’s business model, but also who controls access to information in the years ahead.
Original article
Comments