AI Labs Call for a Slowdown, Then Cut Prices

OpenAI and Anthropic have joined a growing safety debate while launching cheaper AI models for businesses under pressure to control spending. The releases signal that the fiercest near-term contest may be over efficient deployment, not raw capability.
AI Labs Call for a Slowdown, Then Cut Prices

AI Labs Call for a Slowdown, Then Cut Prices
The latest AI race has acquired an awkward dual message: slow down the frontier, but make it far cheaper to use.

The safety argument intensified on Sept. 8, when former Anthropic researcher Jacob Coxon said he had resigned and accused OpenAI and Anthropic of “gambling with our lives.” Anthropic chief executive Dario Amodei subsequently called for an industrywide slowdown in advanced AI development, a position later joined by OpenAI CEO Sam Altman and Elon Musk.

That warning framed Tuesday’s releases from the two labs. Anthropic introduced Claude Opus 5.5, pitching it not as a dramatic leap in capability but as a more economical version of its flagship workhorse. The company says typical workloads can cost roughly 40% less than Opus 5, through lower token prices and fewer tokens required to complete tasks. It is still intended for complex knowledge work and coding — including sensitive areas such as cybersecurity and biology, where safeguards may route flagged requests to an older model.

OpenAI, meanwhile, launched GPT-6 Sol and Luna, lower-cost companions to its more powerful Astra model. Sol is aimed at demanding workloads such as coding; Luna targets high-volume chores including document summaries and information extraction. Both are priced at about half the cost of their predecessors, according to the company.

Altman cast the pricing move as a bid to broaden access, arguing that per-task cost — rather than token cost alone — is the measure that matters. “We want people to be able to use tons of AI,” he wrote, calling widespread use essential to exploring “this new renaissance.”

The commercial logic is stark. Enterprises are experimenting with model routers and cheaper open-weight alternatives rather than sending every job to premium frontier systems. Anthropic’s Dianne Penn said the company was working to make its models’ reasoning and answers “more efficient” so they consume fewer tokens. The result is a market where the safety debate remains unresolved, but customers are already imposing a different kind of slowdown: a budgetary one.

https://foxvector.com

Write a comment