1. AI inference got cheaper
For past couple of weeks most frontier models are bringing their prices down.
OpenAI cut GPT-5.6 Luna pricing by 80%, bringing it down to $0.20 per million input tokens and $1.20 per million output. Its mid-tier Terra model dropped 20%, reaching $2 input and $12 output per million tokens.
Days later, Anthropic also dropped plan to increase price for Claude Sonnet 5, keeping it at $2 per million input tokens and $10 per million output in response to OpenAI’s move. Now just OpenAI and Anthropic , Deepseek also posted their frontier model pricing in the similar range.
The result of this now near-frontier capability is available at a fraction of what it cost even six months ago. This is a big win for most enterprises, as now Agentic workflows that required expensive reasoning models can now be tiered with heavy reasoning at the top, and cheap fast models for everything else.
Atgeir’s take:-
We now see our clients are focused more on “Do we actually know what we are spending across the organisation, and what value we are getting out of it ?” and that’s why LLM observability and cost governance need to be treated as core infrastructure.
We should be aware of which models are being called. How many tokens per workflow are getting spent. What is the cost-per-outcome across use cases.
These are operational basics when inference is cheap enough that consumption grows without anyone actively deciding to grow it.
2. Gemini is now converting your stored procedures.
Another important announcement that happened from Google is that Gemini is now converting your stored procedures. That changes the database migration conversation.
If you ever faced a large scale Oracle or SQL Server migration, you know exactly why it matters.
While schema conversion and bulk data transfer are often automated, teams still face delays when they rewrite years of embedded business logic by hand.
With Database Migration Service now includes AI-assisted code conversion powered by Gemini. By bringing generative AI directly into the migration workflow, teams can convert stored procedures, triggers, and custom functions into PostgreSQL PL/pgSQL code faster and with higher accuracy.
Atgeir’s take:-
As a Google Cloud partner that works across data modernisation and migration engagements, this is a development we have been watching closely.
This “last mile” of database migration, converting procedural logic, has always been the hardest, most expensive, and least predictable phase of any modernisation programme. It is where timelines slip, budgets inflate, and projects lose CxOs confidence.
AI-assisted code conversion does not eliminate the need for experienced database engineers. But it changes the shape of the work. Instead of writing every conversion from scratch, engineers review and refine AI-generated output. The role shifts from translator to reviewer. That is a meaningful compression and not just of time, but of the specialised dual-dialect expertise that has historically been the bottleneck.