Claude Credits You Might Be Missing + An Open-Source PDF Tool Worth a Look

AI and SaaS Roundup · 2026-10-11

Guys, a few things from this week worth knowing if you’re building with AI. There are monthly Claude API credits to claim, a much cheaper Haiku model, and an open-source release that can pull actual data out of charts in your PDFs. 👀

Already paying for Claude Max or Team? There are API credits to claim

This one is easy to miss. Claude Max 5x now includes $100 in API credits each month, and Max 20x includes $200. Team gets $20 per Standard seat and $100 per Premium seat, pooled together with a $500 monthly cap.

You claim them by linking a Claude Console organisation in your billing settings. Your eligible plan needs to have been active for seven days, and Team owners claim the pool for everyone.

These are for building your own apps and agents through the Claude Platform. They don’t cover interactive Claude Code or extra usage in the Claude app. Pro isn’t eligible either. Unused credits expire each billing cycle, and you don’t need to add a payment card to the Platform to use them. If you have auto-reload or purchased credits, usage can continue after the included credits run out.

Announced 7 October.

Source: Anthropic: credit amounts and claiming instructions

Haiku 5.5 is here, and the price difference is pretty big

Anthropic launched Haiku 5.5 on 7 October. For prompts up to 100,000 tokens, the API price is $0.10 per million input tokens and $0.50 per million output tokens. Above that threshold, it’s $0.50 input and $2.50 output, so the prompt size makes a difference.

Anthropic is positioning it for quick, repeated jobs like summaries, classification and smaller tasks inside a larger agent workflow. Think sorting incoming support messages or summarising documents, rather than handing every little job to your most expensive model.

There’s also a Sonnet 5.5 price cut in the same announcement: cache reads have dropped from $0.20 to $0.10 per million tokens. That’s cache reads specifically, not a blanket 50% discount on Sonnet.

Source: Anthropic: Haiku 5.5 announcement and pricing

An open-source model that reads the chart, not just the words around it

LightOn released LightOnOCR-3 on 8 October, and this is a useful one for anyone building around PDFs, receipts or reports.

It can transcribe text, identify where different blocks sit on a page, describe images and extract chart values into tables. The family comes in 0.8B, 1B and 4B sizes, with an Apache 2.0 licence allowing commercial use.

Imagine uploading a report and getting the numbers behind a chart into a comparison tool, or keeping the page layout alongside extracted text so users can jump back to the right section. Those are potential things to build with it, not just another chatbot wrapper.

The team has published the models, code and a demo. You’ll still need compute to run it, and extracted numbers need checking, but there’s something concrete here to play with.

Source: LightOn: models, code and demo