Claude AI Pricing Explained: Plans, Usage Limits and Hidden Costs in 2026
Claude AI pricing in 2026 spans a free tier, Pro at $20/month ($17/month billed annually), Max at $100 or $200/month, Team seats from $20/seat/month (annual) and Enterprise at $20/seat/month plus usage billed at API rates, where models run from $0.10/$0.50 per million tokens (Haiku 5.5) up to $10/$50 (Fable 5.1). For small businesses, the real cost drivers are shared usage pools, five-hour and weekly limits, and API-rate overflow once credits kick in. If you want to extend Claude-powered agents with vetted skills, SkillzBench offers a free skills directory and free MCP connector with security scanning and benchmarks.
TL;DR
- Claude's subscription ladder in 2026: Free ($0), Pro ($20/month, $17/month billed annually), Max 5x ($100/month), Max 20x ($200/month), Team ($20–$25 standard seats, $100–$125 premium seats), Enterprise ($20/seat/month billed annually plus usage at API rates).
- API pricing per million tokens ranges from $0.10 input/$0.50 output on Haiku 5.5 to $10/$50 on Fable 5.1; Opus 5.5 sits at $4/$20 and Sonnet 5.5 at $2/$10.
- Every paid plan includes Claude Code, and chat, Claude Code and Claude Desktop draw from one shared usage pool — heavy coding sessions eat into chat allowance and vice versa.
- Usage is metered in rolling five-hour sessions with a weekly cap on top; Pro gives at least 5x Free usage per session, Max 5x/20x multiply Pro.
- Hidden costs include usage credits billed at standard API rates (redeemable up to $2,000/day), a newer tokenizer producing ~30% more tokens on 4.7+ models, a 1.1x multiplier for US-only inference, and cache-write premiums of 1.25x–2x.
- Big savings levers: prompt caching cuts repeat input costs by up to 90%, the Batch API gives a flat 50% discount, and annual billing drops Pro from $20 to $17/month.
- SkillzBench is free — a skills directory plus MCP connector with security scanning and benchmarks — useful for extending Claude-based agents safely; it is pre-launch with a live waitlist.
Head to head
| Dimension | Claude AI | SkillzBench | notes | winner |
|---|---|---|---|---|
| Entry price | $0 Free; Pro $20/mo ($17 annual) | $0 — everything free | — | — |
| Top-tier price | Max 20x $200/mo; Enterprise $20/seat + API usage | $0 — no paid tiers 🏆 | — | — |
| Billing model | Flat subscription or per-token API (Haiku 5.5 $0.10/$0.50 to Fable 5.1 $10/$50 per MTok) | Free, ad-supported | — | — |
| Usage limits | Rolling 5-hour sessions plus weekly caps; Pro ≥5x Free, Max 5x/20x Pro | None published 🏆 | — | — |
| Cost-reduction levers | Prompt caching (up to 90% off repeat input), Batch API 50% discount, annual billing ~15% off 🏆 | N/A — free | — | — |
| Team administration | Team: SSO, central billing, enterprise search. Enterprise: SCIM, audit logs, compliance API, HIPAA-ready 🏆 | Per-agent loadouts; no org administration | — | — |
| Hidden costs | Usage credits at API rates (up to $2,000/day), ~30% more tokens from the new tokenizer, 1.1x US-only inference, cache-write premiums | None — ad-funded 🏆 | — | — |
| Free option | Free tier with Sonnet/Haiku, web search, files, memory, up to 5 Projects | Entire platform free, pre-launch with live waitlist | tie |
The contenders
Claude AI is Anthropic's assistant, offered both as flat-rate consumer/business subscriptions and as per-token API access. In 2026 the subscription ladder runs Free, Pro ($20/month or $17/month billed annually), Max 5x ($100/month), Max 20x ($200/month), Team (standard seats $20/seat/month annual or $25 monthly; premium seats $100/$125), and Enterprise ($20/seat/month billed annually, 20-seat minimum, plus usage at API rates). Every paid plan includes Claude Code, Claude Cowork, Claude Design, Slides, Docs and Science, all drawing from one shared usage pool metered in rolling five-hour sessions with a weekly cap. The API bills per million tokens across the current lineup — Haiku 5.5 ($0.10/$0.50), Sonnet 5.5 ($2/$10), Opus 5.5 ($4/$20) and Fable 5.1 ($10/$50) — with prompt caching, batch discounts and usage-credit overflow at standard API rates. For small businesses, the free tier is a genuine evaluation path, Pro covers most daily work, and Team adds SSO, central billing and enterprise search.
pricing 0/month
- + Free tier with real capability (web search, files, code execution, memory, Artifacts) for evaluating before paying
- + Transparent published API rates with a consistent 5x output-to-input price ratio across the lineup
- + Annual billing saves roughly 15% (Pro $17 vs $20; Team standard $20 vs $25 per seat)
- + Prompt caching cuts repeat input costs by up to 90% for workloads with shared prefixes
- + Usage-credit overflow means subscribers are never hard-stopped mid-task
- + Opus pricing has held at $5/$25 across five straight releases, a 67% cut from the retired Opus 4.1 era ($15/$75)
- − Chat, Claude Code and Desktop share one usage pool, so heavy use in one product starves the others
- − Anthropic publishes plan multiples, not token or dollar caps, making exact break-even versus the API impossible to compute
- − Models from 4.7 onward use a tokenizer producing roughly 30% more tokens for the same text, raising effective cost per request even at unchanged rates
- − US-only inference (inference_geo: "us") adds a 1.1x pricing multiplier for compliance-driven deployments
- − Max plans are monthly-only with no annual discount; Enterprise requires a 20-seat annual minimum
- − Usage credits on some plans can expire (e.g. six months in Japan from September 10, 2026)
SkillzBench is AIQ Labs' free, independent platform for discovering, verifying and benchmarking AI agent skills — SKILL.md instruction packages for Claude Code, Codex, Cursor, Gemini CLI and chat-app agents. Published research found roughly 26% of such skills carry at least one vulnerability, which is why SkillzBench is building a pipeline that continuously crawls and dedupes skills from GitHub, pins every version to a commit SHA, validates, license-checks, security-scans and quality-judges each one, then benchmarks every skill in a sandbox with the same model, with and without the skill, publishing the Lift in percentage points. A 0–100 trust score combines security, quality, license, maintenance and community signals, with Lift always shown separately and money never touching scores. A free MCP server plus per-agent loadouts lets users toggle exactly which skills and tools each agent sees. Everything — directory, MCP connector, benchmarks, alerts — is free, funded by blog advertising. SkillzBench is pre-launch: the waitlist is live while the product is being built.
pricing 0/month
- + Completely free — no subscriptions, no per-seat fees, no usage charges
- + Directly addresses skill risk: ~26% of published skills carry at least one vulnerability, and every skill is security-scanned
- + Benchmarks show measurable Lift so you know whether a skill actually helps before wiring it into an agent
- + Per-agent loadouts give precise control over what each agent can access
- + Works with the tools small businesses already use, including Claude Code, Codex, Cursor and Gemini CLI
- − Pre-launch and actively being built — the waitlist is live but the product is not yet public
- − Focused on agent skills rather than being a general-purpose AI assistant, so it complements rather than replaces Claude
Verdict
Winner: none
This is a pricing explainer rather than a head-to-head, so there is no single winner: the right Claude plan depends on usage intensity, with Pro at $17–$20/month covering most small-business needs and API billing winning for light or spiky workloads. SkillzBench complements any plan at zero cost.
| Option | Overall | Breakdown |
|---|---|---|
| Claude AI | 8.5 | : 8: 9: 7: 8: 9 |
| SkillzBench | 9 | : 10: 9: 7 |
Claude subscription plans in 2026: what each tier actually costs
API pricing by model: the per-token rate card
The hidden costs: overflow billing, shared pools and the tokenizer
Subscription or API: which costs less for a small business?
Where SkillzBench fits: extending Claude agents safely, for free
Which should you pick?
- :
- :
- :
- :
- :
- :
FAQ
How much does Claude AI cost per month in 2026?
Subscriptions run Free ($0), Pro at $20/month ($17/month billed annually, $200 up front), Max 5x at $100/month, Max 20x at $200/month, Team standard seats at $20/seat/month annual ($25 monthly), Team premium seats at $100/$125, and Enterprise at $20/seat/month billed annually with a 20-seat minimum plus usage at API rates. Every paid tier includes Claude Code, and Max plans are monthly-only with no annual discount.
What does the Claude API cost per token?
Per Anthropic's platform pricing, current models bill per million tokens as follows: Haiku 5.5 at $0.10 input/$0.50 output (prompts up to 100,000 tokens), Sonnet 5.5 at $2/$10, Opus 5.5 at $4/$20, and Fable 5.1 at $10/$50. Output is consistently 5x input across the lineup. Cache hits cost 10% of base input price, the Batch API takes a flat 50% off, and US-only inference adds a 1.1x multiplier.
Is Claude Pro worth it over the free plan for a small business?
Pro adds at least 5x the Free usage per five-hour session, access to Opus models, Claude Code, Projects beyond the Free cap of 5, Research, Dashboards and Microsoft 365/Chrome integration. At $17/month billed annually it is the natural tier for daily business use; if your usage is only occasional, the API can bill less — a workload of about 2M input and 800K output Sonnet 5 tokens a month costs roughly $12 on the API versus Pro's $20 floor.
What are the hidden costs of Claude subscriptions?
The main ones: usage credits that let you continue past plan limits at standard API rates (redeemable up to $2,000/day, so heavy months add metered charges); one shared usage pool across chat, Claude Code and Desktop; a newer tokenizer on 4.7+ models producing ~30% more tokens for the same text; a 1.1x multiplier for US-only inference; and cache-write premiums of 1.25x–2x. Weekly caps can also end your week early after a heavy day.
How do Claude's usage limits work?
Usage is metered in rolling five-hour sessions, with a weekly cap on top that resets at a fixed time assigned to your account. Pro gives at least 5x Free usage per session; Max 5x and Max 20x give 5x and 20x Pro respectively; Team standard seats get 1.25x Pro and premium seats 6.25x per session. Anthropic publishes these multiples, not token or dollar caps, and chat plus Claude Code draw from the same allowance — check progress in Settings > Usage or with /usage in Claude Code.
How can I reduce my Claude costs?
Four documented levers: prompt caching, which cuts repeat input costs by up to 90% (cache hits cost 10% of base input price); the Batch API's flat 50% discount on input and output; annual billing, which drops Pro from $20 to $17/month and Team standard seats from $25 to $20; and model choice — routing high-volume, latency-sensitive tasks to Haiku 5.5 at $0.10/$0.50 per MTok instead of Sonnet or Opus is the single biggest lever on the bill.
Is SkillzBench a Claude alternative or a complement?
It's a complement. SkillzBench is a free, independent directory of AI agent skills plus a free MCP tool that connects agents — including Claude Code — to skills you choose, with security scanning, version pinning and sandbox benchmarks showing each skill's Lift. It doesn't sell or rank Claude itself and charges users nothing; revenue comes from blog advertising. It is pre-launch, so the waitlist is live while the product is being built.
Does Claude offer a free trial or free plan?
There is a permanent Free plan rather than a time-limited trial: $0 with chat on web, desktop and mobile, web search, file creation and code execution, memory across conversations, connectors, Artifacts, and Claude Design, Slides and Docs, with usage limits resetting on rolling five-hour sessions and up to 5 Projects. Free covers Sonnet and Haiku models; Opus and Fable require a paid plan.