What agents really cost — and how to spend the context window like the scarce resource it is.
| Part | Topic | Note |
|---|---|---|
| 01 | What agents really cost | subscriptions, API, Bedrock |
| 02 | Efficient work with context | most important part · live demo |
All pricing and product claims are date-stamped July 2026 and sourced — see the closing Sources slide.
Tokens, priced. One real session's bill, traced line by line. Then: which plan for which team.
Everything sent with the request — prompt, files, tool results, history. The base meter: cheaper than output, pricier than cached.
Everything the model generates — including thinking tokens. The priciest meter, often 5× input.
Repeated context (system prompt, CLAUDE.md, tool schemas) served from cache — up to ~90% cheaper.
July 2026 rates, e.g. Sonnet 5: $3/M input · $15/M output (introductory $2/$10 through Aug 31, 2026). Spot-check platform.claude.com/docs/en/pricing before quoting externally.
Cost and latency are the same lever here — trimming output tokens helps both at once.
The single biggest avoidable cost in most sessions is re-reading the same file because context was already lost or cleared badly — part 02's whole subject.
| Plan | Price | Usage |
|---|---|---|
| Pro | $20/mo (~$17/mo annual) | baseline |
| Max 5× | $100/mo | 5× Pro's limits |
| Max 20× | $200/mo | 20× Pro's limits, extended context, priority support |
No confirmed statement of a new subscription "direction" beyond Pro/Max/Team/Enterprise as of July 2026 — treat any such claim as unconfirmed. Pricing moves periodically; verify at claude.com/pricing before presenting exact numbers.
| Model | Input /M | Output /M | Notes |
|---|---|---|---|
| Fable 5 | $10 | $50 | highest-cost self-serve tier |
| Opus 5 | $5 | $25 | complex agentic coding, enterprise |
| Sonnet 5 | $3 | $15 | $2/$10 intro through Aug 31, 2026 |
| Haiku 4.5 | $1 | $5 | classification, extraction, routing |
Prompt caching: up to ~90% off cached input. Batch API: flat 50% off. Third-party trackers agree here, but spot-check the Anthropic pricing page directly before quoting on stage.
Steady daily interactive coding, one person, usage inside the plan's cap. Predictable monthly bill.
Bursty or programmatic use — CI runs, batch jobs, evals, multi-agent fleets. Pay exactly for what runs.
Hybrid credit-pool models (Cursor's Auto mode — deck 07, Measuring What Works) sit in between: subscription price, usage-aware routing underneath.
Verified SSO paths: Okta, Microsoft Entra ID, Google Workspace.
Cursor claims up to 60% cost savings vs. running everything on Opus — that number is Cursor's own marketing claim, not independently verified.
| Team shape | Reach for |
|---|---|
| Solo engineer, daily interactive coding | Pro or Max, by usage volume |
| CI, batch jobs, scheduled evals | raw API + batch discount (50% off) |
| Regulated data / AWS-committed org | Claude on Bedrock (regional endpoint) |
| Team needing SSO/audit/spend limits | Team or Enterprise |
| Wants routing across model vendors | Cursor Router or OpenRouter (deck 07) |
The most important part of today. Context is the scarce resource — spend it like one.
Three separate failure modes, one fix: be deliberate about what stays in the window.
Everything left of "conversation" is fixed overhead, resent and re-billed every turn (cached, but not free). Only the conversation genuinely grows — and it's the part you control least by default.
Caching makes this cheaper, not free — and a cache miss (from any upstream edit) means the full price, every time. Run /context in Claude Code to see your own session's exact breakdown.
Test: could this live as a code comment or README instead? If yes, it usually shouldn't be here.
Every line in CLAUDE.md is paid for on every turn, of every session, forever. Treat additions like you'd treat a hot-path dependency.
Full reset. Use when the task is genuinely done and the next one is unrelated — don't drag dead context into a fresh task.
Summarize and continue. Use mid-task when the window is filling but you still need the thread.
Wrong choice either direction costs you: /clear too early loses context you needed; /compact too rarely drags a bloated window into every remaining turn. And /compact isn't free — the summarization pass itself reads the whole window one more time.
This is context hygiene applied at the architecture level, not just a discipline you have to remember mid-session.
"Read the whole file just in case" is exactly the re-read pattern that was the biggest avoidable cost line item back in part 01.
Live Watch the token counters, not just the transcripts — the difference is usually bigger than it feels.
Use only if the live run stalls. Narrate the captured turn/token delta from rehearsal, then move on.
If this deck changes one habit, make it this one — it's the highest-leverage change available to most engineers this week.
Run /context before and after: the trim is only worth what the numbers say it's worth.
Three meters, not one — and output (thinking included) is the expensive one. Pick the channel by team shape, not by habit.
The window degrades as it fills, re-bills every turn, and re-caches badly. Trim CLAUDE.md, clear deliberately, read selectively.
Of the two, context is the one to act on first — it's the highest-leverage habit change available this week.
Full source list with per-claim citations: deck05-facts.md (internal research doc). One item flagged unverified: any claim of a new announced subscription direction beyond Pro/Max/Team/Enterprise.