Anthropic's Claude Code Under Fire Amid Drastic Cuts, OpenAI Faces IPO Hurdles

Anthropic is under scrutiny following a series of drastic changes to its Claude Code service, which include an offer of €17 (or equivalent USD) in free credits, seen by many as a conciliatory gesture ahead of further anticipated cutbacks. Users report accelerated credit consumption, with one developer noting 10% usage for a single trivial query. Adding to the friction, Boris Cherney announced that Claude subscriptions no longer cover third-party tools like Open Claude, pushing users towards additional paid usage packs or direct API key use. Anthropic justifies these moves by citing increased demand and the need to prioritize its core products and APIs to ensure sustainable long-term service. While some code corrections were implemented to address rapid token depletion, Anthropic also stated that “peak hours” now entail stricter limits and faster consumption, a change deemed arbitrary and lacking clear user guidance. Official guidance, post-investigation, shifted blame to user-side “large-token patterns,” advising users to switch from Claude 3 Opus to Sonnet for Pro plans, limit context windows, and avoid resuming long inactive sessions. This ambiguity has led to widespread frustration among developers, compounded by Claude Code blocking even its own internal command-line tools for suspected policy violations, effectively dictating how users can interact with their paid subscriptions. These policies have prompted significant user dissatisfaction and a wave of cancellations.

The broader AI landscape also faces internal challenges, as OpenAI’s CFO, Sara Frear, reportedly clashes with CEO Sam Altman over the company’s financial health and an ambitious 2024 IPO target. Frear, a former Goldman Sachs analyst, privately expressed concerns that OpenAI’s revenue growth is insufficient to support its massive spending, making a public offering before 2026 unlikely. This tension highlights the financial pressures, as OpenAI’s full $122 billion funding round is contingent upon a successful IPO, with only $25 billion secured otherwise. In a perplexing move, OpenAI recently acquired TVPN, a daily tech talk show, on April 2, a decision widely viewed as unconventional and lacking clear strategic rationale. These events unfold against the backdrop of staggering projected model training costs, with The Wall Street Journal reporting that OpenAI anticipates spending $125 billion by 2029, and Anthropic around $35 billion, on model training alone. As a creative workaround to Anthropic’s rising costs, a Reddit user demonstrated a “caveman mode” prompt engineering technique, claiming 50-75% savings on output tokens, which are five times more expensive than input tokens, offering a glimpse into the community’s ingenuity in adapting to evolving AI economics.