# Postgres LLM, Vercel AI gateway, SOB Updates, Scraper and more

URL: https://interfaze.ai/blog/06-05-26-updates

Here's what we shipped last week.

## Postgres LLM

Run Interfaze or any LLM right in your Postgres database on row-level changes and store the results across columns.

It's also fully async so it doesn't block your database transactions.

[Github](https://github.com/JigsawStack/postgres-llm) | [Full blog](https://interfaze.ai/blog/run-llms-inside-postgres) | [Demo video](https://www.youtube.com/watch?v=bb6La0GZeBc) | [X Launch](https://x.com/yoeven/status/2051709895649726574) | [LinkedIn Launch](https://www.linkedin.com/posts/yoeven_we-open-sourced-postgres-llm-many-developer-ugcPost-7457477700497317888-9ZBU?rcm=ACoAABYoZGIBzg2ab_Pks_mGonKQKmN6kId1mYo)

![Postgres LLM](https://r2public.jigsawstack.com/interfaze/blogs/postgres_llm_blog_card.png)

## Vercel AI Gateway

We're now on Vercel AI gateway! Now you can swap models easily between deterministic and non-deterministic models. Easily track your usage, billing and observability all in one place along side all your dev tools and LLM usage.

Try now: https://vercel.com/ai-gateway/models/interfaze-beta

![Vercel AI Gateway](https://r2public.jigsawstack.com/interfaze/blogs/vercel_ai_gateway.png)

## The Structured Output Benchmark (SOB)

Last week we launched SOB and we've added Claude Opus 4.7, GPT 5.5, GLM 5.1 and Gemini 3.1 pro.

GLM 5.1 beats Claude Opus 4.7
Gemini 3.1 pro & GPT 5/mini is set to the lowest reasoning since it can't be turned off

Check out the full leaderboard: https://interfaze.ai/leaderboards/structured-output-benchmark

![SOB](https://r2public.jigsawstack.com/interfaze/blogs/sob_benchmark.png)

## Scraper

Learn more: https://interfaze.ai/docs/web/web-search

The scraper now supports scraping pages that besides HTML/Javascript rendered like raw markdown and text. So you can extract a random image url in a large markdown from a Github repo or an email from a text file attached on a website.

## Speech-to-text

Try now: https://interfaze.ai/docs/audio/speech-to-text

Now you can change the per chunk duration of each speaker to any duration just by prompting. It used to default to 3-4s and now you can set it to as low as 1s.

### Coming soon:

- Improved GUI detection speed
- Updated benchmarks on Interfaze
  -Performance improvements on caching
- STT per word timestamps

That's it for this week!

Best,
Yoeven
CEO, interfaze.ai