AI Digest CW38/39: Opus 5.5, Halved Prices, Data Leaks

A look back at calendar weeks 38 and 39 (September 14 to 27, 2026), compiled after the fact from reports by The Decoder, TechCrunch, MIT Technology Review and Anthropic. Six topics relevant to small and medium-sized businesses.

OpenAI halves API prices with GPT-6 Sol and Luna

OpenAI releases GPT-6 Sol and Luna, two cheaper models that are meant to match their predecessors' performance at halved token prices. Independent analyses, however, see hardly any intelligence gains. If you use AI via API, you can review your costs and compare models.

Claude Opus 5.5: faster and, per Anthropic, 40 percent cheaper

Anthropic released Claude Opus 5.5 on September 22, the first model of the 5.5 family, with Sonnet 5.5 and Haiku 5.5 due in the following weeks. According to Anthropic it generates output over 30 percent faster than Opus 5 and makes common workloads 40 percent cheaper. The price is 4 dollars per million input and 20 dollars per million output tokens. All figures come from the vendor.

Exposed Supabase databases: AI-built apps leak user data

According to TechCrunch, some Supabase customers are publicly exposing large amounts of people's data on the web. The findings show that AI-generated and vibe-coded apps can spill user data when they are not configured and secured properly. If you build apps with AI, check access rules and database permissions before going live.

Microsoft Copilot: Autopilot agent and usage-based billing

Microsoft splits the Copilot app into Home, Code and the new Autopilot agent. The agent runs permanently in the cloud and monitors Teams channels, for example. Autopilot and Code are billed by usage instead of a flat rate, which makes costs harder to predict.

Google tests Gemini calls to businesses

Google is testing a feature in which Gemini calls businesses on behalf of users. It will first be available to Pixel 11 owners in the US who pay for a Gemini subscription (TechCrunch). Businesses with a phone service could receive AI callers in the future.

Beware of this summer's AI hype

MIT Technology Review puts the summer's waves of hype in context: at the end of April Anthropic claimed its model Claude Mythos is better at finding software vulnerabilities than most security experts, then came the Hugging Face hack by OpenAI models and similar incidents involving Anthropic and Meta. The piece urges readers not to be fooled. For business decisions, tests in your own setting count.