AI Learning Hub › Your AI Stack
Tool directory
Your AI Stack
Find the right AI tool for the job — then stop. A stack is the small set of tools you actually know how to use, chosen around the work you do, not around whatever launched this month.
113 tools 16 with Stanford-provided access Entries reviewed 10 August 2026
A listing is not an endorsement. A tool appearing here does not mean Stanford Law School licenses it, provides access to it, or approves it for your coursework. Tools marked Stanford access are the exception, and those use Stanford sign-in. Whether AI is permitted for a particular assignment is set by your instructor, and the Honor Code applies either way. Before a new task, walk through the PAUSE Rule.
What an “AI stack” is, and how to build one
Borrowed from software engineering, where a “tech stack” is the set of tools used to build something. Your AI stack is the personal set you combine to do your work: one tool for research, another for drafting, another for the spreadsheet nobody wants to open.
Discover
Search the catalogue by the problem, not the product name. “Transcribe an interview” gets you further than a list of every tool with an AI logo.
Curate
Save the ones worth trying with the ★ button. Start with one or two. A stack of twelve tools you have opened once is not a stack.
Keep it
Your saved list stays in this browser between visits. Download it as a Markdown file or print it to PDF when you want it somewhere else.
Build your stack around the work that matters
Most people get AI wrong in one of two ways.
Some pick one tool and force it into every task, even when the job needs something else. Others chase every launch, bouncing between tools with no system, no depth, and no lasting advantage. Both fail for the same reason: they start from the tool.
The better approach starts from the work — the problems you need to solve, the decisions you need to make, the documents you need to produce, and the standard your role demands. A strong stack is not about access to everything. It is about choosing well, using deliberately, and building a system that makes you faster and sharper over time.
- Do not build your stack around what is trending. Build it around what is useful.
- Do not collect tools. Build capability.
- Do not confuse experimentation with mastery.
Over the next few years the advantage will not go to the people who tried every AI tool. It will go to the ones who got genuinely good with the right few.
The skill areas worth getting good at
Pick one tool in each area that matters to you. Learn it properly. Use it often enough that it changes how you work.
Research and synthesis
Find signal fast, separate quality from noise, and turn a pile of sources into usable insight — with the citations checked.
Writing and communication
Draft, tighten, and clarify across memos, briefs, emails, and proposals without handing over your judgment or your voice.
Analysis and decision support
Break down complex problems, surface tradeoffs, and stress-test an argument before someone else does.
Workflow automation
Remove the repetitive work — the renaming, reformatting, and re-sending that quietly eats a week.
Technical creation
Scripts, quick prototypes, and small tools. No longer a niche skill, and increasingly available to non-programmers.
Data fluency
Work through spreadsheets, dockets, and structured records to get from raw data to a defensible conclusion.
Knowledge management
Capture, organize, and retrieve what you learn so it compounds instead of vanishing into scattered notes and tabs.
Prompting and instruction design
Clear context, sharp inputs, and defined outcomes. The people who get the most from AI are the ones who know how to direct it.
The real goal. You do not need every AI tool. You need the right stack for your work, depth rather than distraction, and real skill in the areas that will still matter once AI stops being novel and starts being expected.
The major general-purpose models
Four labs currently ship general-purpose models at the frontier. Version numbers below were checked against each vendor’s own documentation on 10 August 2026 — and they change every few weeks, so treat the names as a snapshot and the shape of the field as the durable part.
OpenAI — GPT‑5.6 (Sol, Terra, Luna)
Announced 9 July 2026. The number is the generation; Sol, Terra, and Luna are capability tiers — Sol the strongest, Luna the cheapest. OpenAI describes Sol as its strongest model for coding, knowledge work, and science. GPT‑5.5 remains widely deployed while 5.6 rolls out across ChatGPT and the API.
Anthropic — Claude Opus 5 and Fable 5
Claude Fable 5, generally available 9 June 2026, is Anthropic’s most capable widely released model; Claude Opus 5 is the recommended default for complex agentic and professional work, with Sonnet 5 as the speed-and-intelligence balance. The current generation carries a 1M-token context window.
Google — Gemini 3.6 Flash, Gemini Omni
Gemini 3.6 Flash (July 2026) is the workhorse tier, faster and cheaper than 3.5 Flash. Gemini Omni, introduced at I/O in May 2026, generates video from mixed image, audio, video, and text input. The Pro tier is mid-transition: Google said in late July that Gemini 3.5 Pro was still testing with partners, so Gemini 3.1 Pro remains the shipping Pro model.
xAI — Grok 4.5
Released 8 July 2026, with native real-time search and X integration — the distinguishing feature if what you need is what happened this morning. It also carries the fewest content restrictions of the frontier models, which cuts both ways.
Open weights and everything else
Meta moved its frontier work from Llama to the Muse family in 2026: Muse Spark shipped closed, and Muse Glimmer — a 30B multimodal model small enough to run on one GPU — was released under Apache 2.0 on 10 August 2026. Mistral, DeepSeek, and Qwen are worth tracking for open or self-hostable deployments. Stanford users should reach DeepSeek through the Stanford AI Playground (opens in a new tab), which University IT recommends as the Stanford-managed environment for AI exploration.
A version number is not a strategy. The gap between the top four models is smaller than the gap between someone who has learned to direct one of them and someone who has not. Pick the one you have proper access to, learn it well, and re-check this section once a term rather than once a week.
The catalogue
Every tool in the directory, searchable by name, description, tag, or use. Press / anywhere on the page to jump to the search box.
Nothing matches. Try a broader word, clear the category, or search for what you want to do rather than the name of a product.
Tools that have been retired
AI products get shut down, and a workflow built on one that disappears is a real problem. These are products that have closed, been wound down, or have a shutdown scheduled — kept here so a dead link or a stale recommendation can be traced.
Something out of date? Tools rename and disappear faster than any directory can follow. Email library@law.stanford.edu and we will fix the entry.