Ask The PhD Community

HigherEd AI Daily

September 23 – A Cheaper AI Price War Reaches the Classroom

Wednesday, September 23, 2026

Today's developments turn on affordability: cheaper frontier models and open research infrastructure are quietly redrawing which campuses can compete on AI, in the classroom and in the lab.

TLDR — ACCESS

A New AI Price War Could Make Campus AI Budgets Go Further

Anthropic and OpenAI released competing model updates within an hour of each other this week. OpenAI's GPT-6 Sol and GPT-6 Luna launched at roughly half the promotional price of their GPT-5.6 predecessors, while Anthropic's Claude Opus 5.5 dropped its per-token cost below Opus 5.0 while matching the reasoning quality of its Fable 5.1 model across every effort level.

The timing was not a coincidence. Both companies are competing on efficiency as much as raw capability, and token-efficient models translate directly into lower operating costs for anyone running AI at scale, including a university licensing tools across a department or an entire campus.

Why it matters for campuses

For provosts and CIOs weighing AI procurement, falling per-token costs mean the same site license budget now buys meaningfully more capability, or the same capability costs less. That shifts the calculus for under-resourced departments and community colleges that have been priced out of frontier tools, and it is worth revisiting any AI vendor contract signed even six months ago.

Read More

TLDR Dev — RESEARCH

A Blueprint for Universities to Build AI Research Infrastructure Without Big Tech's Budget

Independent AI lab dlab used its open source week to release models, training code, and hardware-conscious tooling aimed at research groups that cannot compete for the largest GPU clusters. The premise behind the release: AI-assisted research has compressed individual projects from years to weeks or days, which makes a single isolated paper a weaker unit of contribution than a shared, reusable research ecosystem.

Rather than treating compute scarcity as a fixed constraint, the release argues that university labs can compete by pooling infrastructure, tooling, and training pipelines the way open source communities already do, instead of chasing hardware parity with industry labs.

Why it matters for campuses

This is a concrete strategy question for research deans and lab directors right now: keep competing for scarce GPU allocations department by department, or invest in shared, campus-wide or consortium-wide AI research infrastructure that smaller labs can actually use. It is also a live equity issue between well-funded research institutions and everyone else.

Read More

TLDR Design — TOOLS

Anthropic Brings Document and Slide Creation Directly Into Claude

Anthropic has launched Claude Docs and Slides, a rich text editor that lets users draft documents and presentations directly from a chat conversation. Claude asks clarifying questions along the way and comments on its own drafting choices rather than simply returning a finished file.

Finished files live in Claude's Artifacts tab and export to Word, PDF, Google Docs, or Markdown; the feature also supports real time collaboration, though Team and Enterprise plans still lack version history. Analysts read the move as an attempt to keep users inside Claude for the whole drafting process, rather than treating Claude as a step before the real work happens in Word or Google Docs.

Why it matters for campuses

For faculty and staff, this collapses drafting and formatting into a single AI-assisted step: a syllabus, rubric, or committee report can be drafted, revised, and exported without leaving the chat interface. The missing version history on Team and Enterprise plans is worth flagging to anyone using it for collaborative departmental documents, since it limits the ability to track who changed what.

Read More

Tool of the Day

LiteParse

LiteParse is a free, open source PDF parsing tool that extracts spatial text with bounding boxes entirely on your own machine, with no proprietary LLM calls and no cloud dependency required. It is built for anyone who needs to pull structured text out of PDFs, forms, and scanned documents at scale, which makes it useful for offices handling large volumes of institutional paperwork.

Try it: Run LiteParse on a folder of scanned syllabi, departmental reports, or course evaluations to pull out structured, searchable text, without uploading student or personnel records to a third party cloud service.

Visit LiteParse

Have a great learning day!

Dr. Ali Green

Sources for This Edition

TLDR (tldrnewsletter.com)
TLDR Dev (tldrnewsletter.com)
TLDR Design (tldrnewsletter.com)
Simon Willison (simonwillison.net)
Tim Dettmers / dlab (timdettmers.com)
Computerworld (computerworld.com)
LiteParse / run-llama (github.com)

askthephd.com
 | 
askthephd.substack.com
 | 
Unsubscribe

HigherEd AI Daily; Curated by Dr. Ali Green