What Claude actually costs.
Claude has a free tier, a $20 Pro plan, two pricier Max tiers, and a per-token API with three model sizes. Here's what each costs and what a real prompt runs you — honest September 2026 numbers.
01 The plans
| Plan | Price | Who it's for |
|---|---|---|
| Free | $0 | Trying Claude, light daily use |
| Pro | $20/mo | Most users — solid limits, Projects, Artifacts |
| Max (5×) | $100/mo | Heavy users wanting 5× the usage |
| Max (20×) | $200/mo | Power users living in Claude all day |
In the app, you're not billed per word — you get a usage allowance measured in messages, and the Max tiers simply raise that ceiling. Most people are well served by Pro.
02 Pay-per-use (the API)
Claude's API charges per token (~¾ of a word), and you pick a model size by how hard the job is. Current rates, per million tokens (input / output):
| Model | Input / Output (per 1M) | Use it for |
|---|---|---|
| Claude Haiku 5.5 (Oct 7, 2026) | $0.10 / $0.50 (up to 100K input tokens; $0.50 / $2.50 above that) | Fast, cheap, high-volume tasks — summaries, classification, subagents |
| Claude Haiku 4.5 | $1 / $5 | The previous Haiku |
| Claude Sonnet 5.5 (Sep 28, 2026; Sonnet 5 is the same price) | $2 / $10 — Sonnet 5 pricing was made permanent Aug 10, 2026 | The everyday workhorse |
| Claude Opus 5.5 | $4 / $20 — cache reads $0.20 | Fable 5.1-level on most work, per Anthropic (released Sept 22, 2026) |
| Claude Opus 5 | $5 / $25 | The previous Opus (released July 24, 2026) |
| Claude Fable 5.1 | $10 / $50 | The top tier — long, complex, agentic jobs (released Sept 1, 2026) |
Fable 5.1 update · September 1, 2026: the per-token rates above did not move, but Fable 5.1 cut its cache read price from $1 to $0.25 per million tokens — 2.5% of the input rate, where most other Claude models charge 10% (Opus 5.5, below, charges 5%). Cache reads are what you pay when a request re-uses context the model already processed, so this only helps if you send the same long context repeatedly. Anthropic puts the saving at roughly 25% on typical workloads and up to 45% on heavily agentic ones. Nothing changes if you use Claude in the app — this is an API-side price. Full explainer: Claude Fable 5, explained.
Opus 5.5 update · September 22, 2026: the new Opus is cheaper than the one it follows — $4 / $20 per million tokens against Opus 5’s $5 / $25, with cache reads at $0.20, now the lowest cache-read price of any Claude model. Anthropic says it performs at Fable 5.1’s level on most work and costs about 40% less per task than Opus 5, because it also uses fewer tokens. On subscriptions, Anthropic is raising five-hour usage limits on Pro, Max, Team and seat-based Enterprise plans and giving subscribers a one-off limit reset (Settings → Usage, on web or desktop). What it means for your model choice: Claude Opus 5.5, explained.
Haiku 5.5 update · October 7, 2026: Anthropic's new small model costs $0.10 / $0.50 per million tokens for prompts up to 100K tokens (the model has a 1M-token window, so longer prompts bill at $0.50 / $2.50) — Anthropic says about 75% less than Haiku 4.5 on average. Cache reads are $0.01 per million, and the Batch API takes another 50% off. Two caveats from Anthropic's own docs: the newer tokenizer counts the same text as roughly 30% more tokens than Haiku 4.5 did, so your real saving is smaller than the sticker price, and Anthropic says Sonnet 5.5 and Opus 5.5 remain the better choice for complex agentic coding. Sonnet 5.5's cache-read price was also cut to $0.10.
Picking the right size is the whole cost game: Fable is 100× Haiku 5.5 on input and Opus 5.5 is 40×. Use the small model for routine work and save Opus for jobs that truly need it.
03 What a prompt actually costs
- Quick question (~250 in / 350 out): a fraction of a cent on any model.
- Summarize a 2-page doc (~800 in / 1,200 out): about $0.014 on Sonnet 5 or 5.5 (same $2/$10 rate) — and that rate is permanent as of August 10, 2026; about $0.001 on Haiku 5.5.
- Deep analysis on Opus 5.5 (~2,500 in / 3,500 out): roughly $0.08 (it was ~$0.10 on Opus 5) — the premium model on a long job is where cost shows up.
Claude's writing and reasoning quality is its draw — and for normal use, prompts are pennies. The cost lever is model choice: don't run everything on Opus. For most daily work, Pro ($20) flat beats per-token, and Sonnet is the right default if you're on the API.
04 Which is right for you?
- Trying it / light use: Free tier.
- Daily driver: Pro ($20).
- All-day heavy use: Max ($100 or $200) — only if you keep hitting limits.
- Builder: API — default to Sonnet, drop to Haiku for volume, reach for Opus sparingly.
Compare Claude against every other tool — subscription vs pay-per-use — with your real numbers.
Open the AI Cost Calculator →Quick gut-check
If you're on the API, look at how often you reach for Opus. Most jobs run fine on Sonnet at half the price of Opus 5.5 (and a fifth of Fable). Reserving Opus for the genuinely hard ones can cut your bill dramatically without hurting output.
Run the numbers →What you can do now
- Know the plans: Free, Pro ($20), Max ($100 / $200)
- Read the API rates by model: Haiku 5.5 $0.10/$0.50 (Haiku 4.5 $1/$5), Sonnet 5.5 $2/$10 (same as Sonnet 5, now permanent), Opus 5.5 $4/$20, Fable 5.1 $10/$50
- If you send the same long context over and over on the API, Opus 5.5’s $0.20 cache-read rate is now the cheapest of any Claude model (Fable 5.1: $0.25)
- Control cost with model choice — Sonnet by default, Opus sparingly
- Pick the plan that matches your real usage