Source: raw/Every_Claude_2.0_Concept_Explained_for_Normal_People.md; plan facts checked against ai-research/anthropic-support-what-is-the-ultra-plan-2026-11-13.md (Claude Help Center, first-party)
Creator: Nate Herk (self-identifies via “you are Nate Herk’s AI operating system”; captioned “Nate Herks”) · URL: https://www.youtube.com/watch?v=tcb5q_6jwvY · Duration: not stated in the raw file · Platform: YouTube (auto-captions, fetched 2026-11-13)
A beginner’s map of the Claude product as of November 2026, in five parts: the models, where you use Claude, what it makes, what it knows about you, and how it works on its own. “Claude 2.0” in the title is Nate’s nickname for the last 12 weeks of releases, “obviously a nickname, not a new model.” Nearly every concept already has its own article in this wiki; this page is the index to them, plus Nate’s own measurements and usage habits, which are the parts not found elsewhere.
Key Takeaways
- Nate’s own numbers (not found elsewhere in the wiki):
- Sonnet against Opus on seven of his workflows: Sonnet won clear-instruction tasks (a resource guide in “about half the time” and usually cheaper); Opus won vague ones (a 90-day plan).
- Across seven tasks Claude read about 160 million tokens and wrote about 800,000, roughly 200 read per token written. His conclusion: the files and instructions you load matter more than prompt length.
- Same prompt to Opus 5.5 at every effort level:
lowabout 17 minutes, under 26;maxabout 2.5 hours, about $50, still buggy. - He keeps
mediumas his default, usesxhighfor video editing, and suggests Sonnet 5.5 atmediumwith skills for docs and general knowledge work. - He has more than 150 auto-memory notes, more than 100 skills and seven mods. About 20% of everything he sends is a
/goal.
- Plans, as he gives them, match Anthropic’s help page: Free 20, Max 200 plan now called Ultra: 10x Pro’s weekly usage, no five-hour limit for verified accounts, and monthly API credits. Session and weekly limits are shared between the Claude app and Claude Code. Check with
/usage. See Claude Usage Limits Playbook. - Models. Haiku, Sonnet and Opus are all on 5.5. Opus is not on the free plan. On Max and Ultra, Fable can use up to half the weekly limit; on Pro it runs on usage credits (matches Fable 5.2).
/advisorlets a session consult a stronger model at key moments (Advisor Tool). - Three tabs. Chat (docs, slides and designs inside any conversation; web search, memory, connectors and artifacts on the free plan), Code (Claude Code, Pro and up), and Cowork, which Anthropic describes as “Claude code for the rest of your work.” Nate’s advice is to learn Claude Code instead of Cowork, since “eventually you might need the power of code.”
- Sessions and side chats.
Cmd+Nopens a session, split view shows two, and Claude can message and manage other sessions.Cmd+;or/btwopens a side chat that sees the session but adds nothing back to it; the desktop app does not save side chats. See The Revamped Claude Desktop App. - Claude Duet is Anthropic’s new agent app: you message it like a person, it works in connected apps, and asks before sending, paying, signing or deleting. Research preview, waitlist, US only. See Claude Duet Overview.
- What it makes. Artifacts start private. Docs, slides and design have been out of beta on every plan since 2026-10-08 (45 million made); the standalone
claude.ai/designsite closes in December. Dashboards (beta, paid plans) show the query behind every number. Motion (beta, Team and Enterprise) builds animation in code and exports MP4. See Claude Dashboards and Motion. - What it knows. Auto memory notes in Claude Code; CLAUDE.md, which he keeps as a routing map of where things live and under about 200 lines; skills, which load only a short description until needed. His rule: “if you’ve ever done something twice, turn it into a skill.”
- Connectors, plugins, mods. Connectors (many over MCP) work on the free plan. Plugins bundle skills, commands, connectors and agents; his team runs a branded-doc plugin for any external document. Mods change Claude Code itself; you can describe one and Claude builds it. His include a session-cache keeper, a goal progress bar, a pixel cat that announces finished sessions, and a warning when two sessions are about to edit the same file.
- Working on its own. Five permission modes (manual, accept edits, plan, auto, bypass); he leaves auto mode on. Subagents work in their own context and return a summary; his Roast Room council skill runs each board seat as a custom subagent. Claude Code projects start a cloud thread per task (beta on Pro, Max and Ultra).
/goalkeeps working until a stated condition is met. Scheduled tasks run on the desktop; routines run in the cloud on a schedule or on events.
Try It
- Start with three things, in Nate’s order: a lean CLAUDE.md, one skill for the task you do most, then a few
/goalprompts with an unambiguous finish line (“so there’s no question about when it’s done”). - Run your own effort test before raising effort: give the same prompt at
mediumandxhighand compare time, cost and result. Usemaxonly ifxhighfalls short. - Find skill candidates by listing requests you have typed more than twice. His example: “process this YouTube video into the Google Sheet master database and the wiki” became a
process videoskill after about 15 repeats. - Check your plan before assuming a feature is missing. Many features in the video are plan-gated or waitlisted.
Open Questions
- Fable version and price ratio. He calls Fable 5.1 the top-tier model and says it costs “two and a half times what Opus does.” That matches Fable 5.1 (50 against Opus 5.5’s 20). Fable 5.2, released 2026-11-09 at 40, is 2x Opus 5.5. The video probably predates 5.2 or did not update this line.
- “Opus 5.5 performs near Fable level.” Anthropic’s wording was Fable 5.1 level “on most work” (Opus 5.5).
- 1M tokens “roughly 555,000 words” is attributed to Anthropic; not checked.
- “Ultra Code” is described as a setting where Claude plans the whole workflow before starting. Elsewhere in the wiki
ultracodeis the dynamic-workflow keyword that orchestrates many agents; his description may be a simplification. - Auto mode as the default “on Pro, Max, and Team plans since August.” The wiki records auto mode as the default from 2026-08-14 and the universal default from v2.1.284; the plan scoping is his.
- API credits on Max. He says 200 on Ultra; the help page snippet says Ultra’s 100 on Max”. The Ultra help page itself does not state the amount.
- His government remark (“the government actually took away” Fable 5 at launch) is not explained in the video. See Mythos 5 Federal Shutdown for what the wiki has.
Related
- Every Level of Claude (Nate Herk) — his earlier five-level version of the same overview
- Claude Usage Limits Playbook — session and weekly limits in detail
- The Revamped Claude Desktop App — sessions, split view, side chats
- Claude Dashboards and Motion — the October artifact types
- Claude Mods — changing Claude Code itself
- goal Walkthrough — the loop he uses for 20% of his prompts
- Claude Code Projects — cloud threads per task
- Claude Duet Overview — the agent app