What changed
Anthropic released Claude Sonnet 5.5 this week, keeping the same $2 and $10 per million token pricing, but output is now about 30% faster and up to 30% cheaper per task. VentureBeat and CNBC both reported that the new model gets the same work done with fewer tool calls and tokens, so every AI-powered workflow costs less to run. ElevenLabs released v4, which now supports 90 languages, can clone a voice in 10 seconds, and has lower latency for real-time voice agents.
At the same time, OpenAI paused its most advanced model training and shelved GPT-6.1 Astra over safety failures, as reported by CNBC and AI Weekly. This means the most reliable and cost-effective AI stacks for small businesses are now built on Anthropic and existing models, not waiting for the next OpenAI release. The cost of running AI agents to handle everyday front-desk and back-office tasks is now low enough that it is a routine business expense, not a special project.