Anthropic Will Charge Extra for Claude: The End of Flat AI Subscriptions

What Happened?

Anthropic, the company founded by former OpenAI members, has announced that starting now, Claude subscribers (both Pro and Team plans) will have to pay additional usage fees to access its most advanced AI model, Claude Fable 5. According to Wired, the company will implement a system of inference credits that will be consumed based on query complexity. Users who exceed their monthly quota included in the subscription will need to purchase additional packages.

This move represents a radical shift in Anthropic's pricing strategy. Until now, the company offered unlimited access to its most powerful models within subscription plans, following the trend set by OpenAI with ChatGPT Plus ($20 per month) and Google with Gemini Advanced ($22.99 per month). However, the computational costs of frontier models have grown exponentially. According to industry estimates collected by The Information, the inference cost of Claude Fable 5 could be 10 to 100 times higher than that of Claude 3 Opus, depending on the task. This model, launched in March 2025, has been designed for complex reasoning tasks, code generation, and analysis of large volumes of data, requiring immense computational capacity.

Anthropic has not revealed the exact per-credit fees, but sources close to the company indicate that Pro plan users ($20 per month) will receive a limited number of credits, while Team plan users ($25 per month per user) will have a larger allocation. Once exhausted, users can purchase additional credit packages, the price of which has not yet been announced. This model resembles the token system already used by Anthropic's APIs, but now it is being extended to the end consumer.

Why Is This Important?

Until now, major AI assistants (ChatGPT Plus, Gemini Advanced, Claude Pro) offered unlimited access to their most powerful models for a fixed monthly fee. This business model, while attractive to users, has proven unsustainable due to high computational costs. Anthropic is the first major player to break this trend, which could mark the beginning of an era where usage costs are passed directly to the consumer.

Anthropic's decision is not surprising given the context: the training and inference costs of frontier models have skyrocketed. According to a report by SemiAnalysis, training Claude Fable 5 required an estimated investment of between $500 million and $1 billion, and daily inference operating costs can reach millions of dollars. To put it in perspective, running a single complex query on Fable 5 can cost up to $0.50 in computational resources, while on previous models like GPT-4 it cost around $0.01. This makes the flat subscription model unviable if a significant fraction of users perform heavy queries.

Historically, the software industry has undergone similar transitions. In the 1990s, enterprise software companies charged perpetual licenses, but with the advent of the cloud, they migrated to monthly subscriptions. Now, with AI, the variable cost of computing is driving a new consumption-based pricing model. Anthropic is leading this transition, but it is not alone. OpenAI has already experimented with message limits on ChatGPT Plus (50 messages every 3 hours) and has hinted in its developer documentation that it could introduce pay-per-use models in the future. Google, for its part, has limited access to Gemini Advanced for certain intensive features.

Market Consequences

This change could have several repercussions:

• For users: Current subscribers will see the value of their subscription reduced if they rely on the most advanced model. Many are likely to reconsider whether the additional cost justifies the benefit, especially compared to alternatives like ChatGPT Plus that still maintain the flat model. However, light users (short queries, simple tasks) will barely notice the difference, as their credit consumption will be low. Anthropic could segment the market: users who need computing power will pay more, while occasional users will benefit from lower prices. This could be positive for service sustainability but negative for perceived value.
• For businesses: Companies that integrate Claude via API were already accustomed to pay-per-use, but Team plans will now also be subject to limits. This could incentivize businesses to seek cheaper providers or optimize their queries. For example, a company using Claude for extensive document analysis could see its monthly costs rise from $25 per user to hundreds of dollars if consumption is not managed. This could accelerate the adoption of smaller, specialized models for routine tasks, reserving frontier models for critical tasks.
• For competition: If Anthropic's move succeeds, OpenAI, Google, and others are likely to follow suit. In fact, OpenAI has already hinted at introducing pay-per-use models in the future, and Google has begun limiting access to Gemini Ultra. This would accelerate the transition toward an AI ecosystem where unlimited access is a luxury. However, it could also open opportunities for new players offering flat subscription models with lower-capability models, capturing price-sensitive users.

Additionally, this change could impact regulation. European regulators, who are already investigating the pricing practices of big tech companies, could examine whether these credit models are transparent and fair. The European Commission has shown interest in ensuring that consumers are not misled by complex credit systems that hide the true cost.

What Should Readers Know?

Claude subscribers should carefully review the new terms of service. Anthropic has not yet published the exact fees, but it is expected that credits will be consumed faster for complex tasks such as code generation, long document analysis, or multi-step reasoning. It is not confirmed whether free plans will be affected, although it is likely they will be further limited. Currently, the free Claude plan allows a limited number of messages per day, and Anthropic may reduce that limit further to incentivize paid subscriptions.

For those who use Claude primarily for light tasks (short chats, simple questions), the impact could be minimal. However, power users will need to assess whether the additional cost justifies the superior quality of Fable 5 compared to other models. For example, if a developer uses Claude to generate code for 8 hours a day, they could consume hundreds of credits daily, which could translate into tens of additional dollars per month. In that case, it might be more cost-effective to use Anthropic's API directly, which already charges per token, or switch to a model like GPT-4o, which does not yet have additional usage fees.

Ultimately, Anthropic's announcement is a symptom that the AI industry is maturing: the days of flat subscriptions with unlimited access to cutting-edge artificial intelligence are numbered. The question now is how long it will take for others to follow suit. In the meantime, users will have to adapt to a new paradigm where the power of AI has a variable price, similar to how we pay for electricity or bandwidth. This change may be painful in the short term, but in the long term, it could allow AI companies to invest in even more powerful and sustainable models.