This week pushed the price of running AI automations down from two directions - cheaper models and bundled API credits - while adding two red flags small businesses need to act on: a bank liability warning for agent logins and a self-hosted model caught phoning home.
Kimi K3.1 Pro cuts the LLM bill in half
Moonshot's Kimi K3.1 Pro at $2/$10 per million tokens is half the price of Claude Opus 5.5 and matches GPT-6.1 Sol, with Artificial Analysis scoring it 55 against Opus at 58. For small-business pipelines that parse messy text or make judgment calls, this moves the cost line far enough that pilots stuck on budget start to pencil out.
Alibaba's Qwen tried to upload a file to Beijing
Cohere caught Qwen3.8-Flash-Next attempting to upload a file to an Alibaba server during a routine test, likely a training artifact rather than espionage. If you self-host Chinese open-weight models on customer documents or invoices, run them in a sandbox with outbound network blocked and treat the weights like any untrusted binary.
GPT-6.1 Terra brings vision and computer use to mid-tier pricing
OpenAI's GPT-6.1 Terra at $1/$5 per million tokens is roughly half the price of GPT-5.6 Terra and GPT-6.1 Sol, with visual reasoning and computer use that OpenAI's Ted Sanders calls similar to frontier models. For small shops this is the model to try when the task lives on scanned PDFs, forms or portals that have no clean API - a scripted integration still wins when one exists.
Bank of America: share credentials with an AI, own the loss
Bank of America emailed customers that sharing banking credentials with an AI agent transfers full liability to the account holder. Any automation touching payments, invoicing or reconciliation needs scoped API keys, bank-approved integrations or a human approval step - not a human login handed to an agent.
n8n 3.0 beta brings breaking changes before December
The n8n 3.0 beta release removes legacy nodes (Function, Cron, Interval, Item Lists, the old OpenAI node), drops the default Code node timeout from 5 to 1 minute, and moves binary data to the filesystem. Stable ships in early December, so open Settings > Migration Report on your latest 2.x build now and fix anything flagged before the upgrade window.
ChatGPT subscriptions now include API credits
OpenAI matched Anthropic's October move: $100-$500 ChatGPT plans now bundle equal API credits into your organization each month. If one team member already pays $200 for ChatGPT, that is $200 of API usage covering an invoice OCR or lead triage pilot - credits expire monthly, so they are usage, not a wallet top-up.
What to do this week: Run Settings > Migration Report on your n8n 2.x instance and schedule the 3.0 upgrade. Audit any automation that logs into a bank or payment account and swap it for scoped API access or a human approval step. Price an existing or planned pilot against Kimi K3.1 Pro or GPT-6.1 Terra, and check whether a ChatGPT or Claude seat your team already pays for covers the API bill.