Reference

Reference material

Use selectively

Product details and supporting examples from the original September 2026 deck.

Verify before reuse

Check availability, prices and limits with the linked primary sources.

Return to the main talk

Core · cheat sheet

Words you'll keep hearing

TokenA chunk of text ≈ ¾ of a word (un·believ·able = 3). What models read & bill in.

PromptThe text you send the model.

ContextEverything the model sees in one call.

Context windowThe max size of that context (1M tokens on today's top models).

Tool callAn action the model asks the harness to run.

HarnessThe program running the model + its tools.

AgentModel + harness, doing things in a loop.

SOTA"State of the art" — today's best.

MCPModel Context Protocol — a standard plug for tools/data.

Don't worry about memorising these — every one gets a slide of its own. This is just your map.

Landscape · models

Today's best models

Maker	Flagship lineup (September 2026)	Known for
Anthropic	Claude Fable 5.1 · Opus 5.5 · Sonnet 5 · Haiku 4.5	coding, agents, long context
OpenAI	GPT‑6 — Astra · Sol · Luna	general reasoning, Codex
Google	Gemini 3.8 Flash · 3.5 Flash‑Lite (3.5 Pro still unreleased)	multimodal, large context

The same three tiers repeat everywhere: large (smartest, priciest) · medium (the sweet spot) · small (fast & cheap) — and Anthropic adds a frontier tier on top (Claude Fable 5.1). It reshuffles every few months, so treat any list as a snapshot. How fast does it move? —

96%

SWE-bench Verified — Opus 5 fixing real bugs in real code projects

2× the field

ARC-AGI-3 — GPT‑6 Astra scores 63%, twice the next-best model, on puzzles built to resist memorising

~6 months

how long "what AI can't do" assumptions stay true

Landscape · picking a tier

Which tier when?

Large · Opus

Hard reasoning, tricky multi-file refactors, architecture. Pricier per word — but on hard work it gets there in fewer steps.

Medium · Sonnet

The daily driver. Most coding & agent work. Often beats last gen's flagship at a fraction of the cost.

Small · Haiku

High-volume, simple, fast: classification, quick edits, sub-agent grunt work.

Example: a cheap Haiku sub-agent skims 200 files and summarises; the pricey Opus only does the actual design call. Cheap legwork, expensive thinking only where needed. But "pricey" is per word, not per job — that turns out to matter, and we come back to it in a few slides.

Landscape · harnesses

Today's best harnesses

Claude Code

Anthropic. CLI + IDE + web. Agentic coding over your whole repo.

Codex

OpenAI's coding agent on GPT‑6 — in your terminal, with a cloud option.

Cursor

An IDE built around an AI agent.

GitHub Copilot

Agent + autocomplete inside your editor.

Devin Desktop

Another agentic IDE (ex‑Windsurf).

Aider

Open-source terminal coding agent.

The harness decides what the model can actually do — tools, sandboxing, memory, parallelism. At least as important as the model.

Landscape · beyond code

Lovable: describe an app, get an app

Lovable's homepage: 'Build something Lovable — create apps and websites by chatting with AI', with a single prompt box

You describe the app in plain language — Lovable builds a working web app: pages, database, login, hosting.
Then you iterate by chat: "make the header sticky", "add a monthly chart" — it edits the live app.
Underneath: the same agent loop we just saw, wrapped in a website. Same category: v0, Bolt, Replit.

Example: "Build an expense tracker: add expenses, a monthly chart, login." → minutes later, a running app on a shareable link — source code included.

Landscape · beyond code

Higgsfield: video from a prompt

Text or image in → finished video out: cinematic camera moves, characters that stay consistent across shots.
An AI director splits your idea into shots, generates each, stitches them — our loop again, different domain.
Built for marketing & social — even "product URL in → ready-made ad out". Same category: Veo, Kling, Runway.

Higgsfield's explore page with tiles for Cinema Studio, Seedance 2.0, and other video generation tools

Example: "A 15-second ad for our mobile bank: phone in hand, city at dusk, upbeat." → storyboard → shots → rendered clip, ready for social.

Landscape · beyond code

HyperFrames: agents that make video

HyperFrames homepage: 'HyperFrames lets AI agents compose videos by writing HTML, CSS and JS' — an HTML file on the left rendering to a video on the right

Open-source (HeyGen): a web page with timing attributes in → a pixel-perfect MP4 out.
Built for agents: Claude Code installs it as a skill, then edits video by chat — the way it edits code.
Same file in → same video out, every time — ideal for explainers, product demos, captions.

Example: "Turn this slide's diagram into a 20-second animated explainer." → the agent writes the HTML, renders it, hands you the MP4.

Landscape · plans

Subscriptions (consumer)

Tier	Claude (Anthropic)	ChatGPT (OpenAI)
Free	limited usage	limited usage
Entry	Pro — ~$20/mo	Go ~$8 · Plus ~$20/mo
Power	Max 5× ~$100 · Max 20× ~$200/mo	Pro 5× ~$100 · Pro 20× ~$200/mo (20× sign-ups paused)

Higher tiers buy more usage and priority access to the newest models. Prices as of September 2026 — they change often, so verify before quoting.

Reference · task cost

Compare the cost of an accepted result

Cheap tokens, many retries

A low unit price can be offset by more tool calls, reasoning and failed attempts.

Higher unit price, fewer attempts

A more capable model can win on a difficult task. Measure rather than assuming it will.

Compare the same task and acceptance check. Include all attempts, cached input, fresh input and output.

Reference · permissions

Isolation must restrict real access

Use a disposable environment with only the files the task needs.
Limit network destinations and remove production credentials.
Apply a time or token budget and retain the final diff and logs.
A project copy, worktree or CI runner is not automatically a security boundary.

Keep approvals where the environment cannot safely contain the consequence.

Sources · further reading

Sources & links (1/2)

Models & pricing

Anthropic — Claude models overviewplatform.claude.com/docs · models Claude — plans & pricingclaude.com/pricing Anthropic — Claude Opus 5 (benchmarks)anthropic.com/news · claude-opus-5 Anthropic — Opus 4.5: effort & token efficiencyanthropic.com/news · claude-opus-4-5 OpenAI — GPT-6 (Astra · Sol · Luna)openai.com/index/gpt-6-astra ARC Prize — ARC-AGI-3 leaderboardarcprize.org/leaderboard ChatGPT — pricing tierschatgpt.com/pricing Google — Gemini 3.8 Flashblog.google · 3-8-flash

Harnesses

Claude Code — memory & CLAUDE.mdcode.claude.com/docs · memory Claude Code — sub-agentscode.claude.com/docs · sub-agents OpenAI — Codexopenai.com/codex Codex CLI — sourcegithub.com/openai/codex

Sources · context, memory & MCP

Sources & links (2/2)

Context, caching & reasoning

Anthropic — Effective context engineeringanthropic.com/engineering Chroma — Context Rot (18-model study)trychroma.com/research/context-rot Liu et al. — Lost in the Middlearxiv.org/abs/2307.03172 Anthropic — Prompt cachingplatform.claude.com/docs · prompt-caching Anthropic — Extended / adaptive thinkingplatform.claude.com/docs · thinking Anthropic — Reasoning models don't always say what they thinkanthropic.com/research · cot-faithfulness

Skills, plugins & MCP

Anthropic — Agent Skillsanthropic.com/engineering · agent-skills Anthropic — Skills repo (SKILL.md)github.com/anthropics/skills Anthropic — Introducing MCPanthropic.com/news · model-context-protocol Model Context Protocol — spec & docsmodelcontextprotocol.io

All figures verified September 2026. Models & prices move fast — re-check the primary sources before quoting.