From last week's radar
What a brief actually looks like
Short, sourced, and written to be acted on in minutes.
Act soonDeprecationOpenAI
Two legacy GPT models retire in 90 days
OpenAI published shutdown dates for two legacy chat models. API calls to those model names will start returning errors after the cutoff.
Next step: Search your codebase and no-code tools for the retired model IDs and switch to a current model before the cutoff.
Source: OpenAI deprecations pageAct soonSecurityMicrosoft 365 Copilot
New admin control for Copilot web search grounding
Admins can now disable web grounding for Copilot per user group. The setting ships off-by-default for new tenants and on for most existing ones.
Next step: Decide whether web grounding fits your data policy, then set the policy explicitly instead of inheriting the default.
Source: Microsoft Copilot data & privacy docsWorth notingPricingAnthropic
Batch API discount widened to more model tiers
The 50% batch-processing discount now covers additional Claude models, lowering costs for overnight or non-urgent workloads.
Next step: Move any delay-tolerant workloads to the Batch API and re-check your monthly run-rate.
Source: Anthropic pricing pageRead the full sample brief