Claude’s pricing has two distinct tracks: subscription plans for individual and team use, and API token costs for developers building with the models directly. Mixing them up leads to bad budgeting decisions. This guide covers both tracks, what each plan actually unlocks, what you pay per token on the API, and where the real cost multipliers are — including Claude Code and agentic usage.
Claude Subscription Plans
Anthropic offers five tiers for accessing Claude through claude.ai. Each plan sets a usage ceiling for the chat interface and Claude Code — API access is billed separately.
| Plan | Monthly Price | Annual (per month) | Who It’s For |
|---|---|---|---|
| Free | $0 | $0 | Light personal use |
| Pro | $20 | ~$17 | Regular users, developers |
| Max 5x | $100 | — | Heavy daily users |
| Max 20x | $200 | — | Power users, agentic workflows |
| Team Standard | $25/seat | ~$20/seat | Small to mid-size teams |
| Team Premium | $125/seat | ~$100/seat | Engineering teams needing Claude Code |
| Enterprise | Custom | Custom | Large orgs, compliance requirements |
Free
The free plan gives limited access to Claude through the web interface, with no Claude Code and no API access. Good for occasional use; not a serious option for development or daily productivity work.
Pro ($20/month)
Pro gives 5x more usage than Free and unlocks access to all current model tiers — Haiku 4.5, Sonnet 4.6, and Opus 4.7. Claude Code is included. The $17/month annual rate makes it the most cost-effective entry point for individual developers or anyone who uses Claude as a primary tool.
Max Plans ($100–$200/month)
The Max tiers exist because Claude Code and extended agentic sessions consume usage allowances fast — roughly 5 to 20 times faster than conversational chat. If you hit Pro limits regularly on coding tasks, Max 5x ($100) covers 5x Pro’s usage. Max 20x ($200) is for developers who run multi-step agents or long coding sessions throughout the day.
Team Plans ($25–$125/seat/month)
Team Standard suits organizations that want shared workspace, admin controls, and standard model access without Claude Code requirements. Team Premium adds Claude Code access for engineering teams and is the practical team-level equivalent of a Max plan. Minimum seat counts may apply — check the current Anthropic pricing page.
Enterprise
Enterprise pricing is custom and typically includes dedicated support, advanced security and compliance features (SSO, audit logs), and negotiated rate caps. Base rates generally start around $20/seat plus API usage, but actual terms depend on contract size and requirements.
Claude API Pricing by Model
API access is pay-as-you-go, billed per million tokens (MTok). Input and output tokens are priced separately. As of July 2026, the active model tiers are:
| Model | Input ($/MTok) | Output ($/MTok) | Context Window | Max Output |
|---|---|---|---|---|
| Claude Opus 4.7 | $5.00 | $25.00 | 1M tokens | 128K tokens |
| Claude Sonnet 4.6 | $3.00 | $15.00 | 1M tokens | 128K tokens |
| Claude Haiku 4.5 | $1.00 | $5.00 | 200K tokens | 64K tokens |
A consistent pattern: output tokens cost 5× the input rate on every tier. That ratio matters for cost estimation — a workflow that generates long completions will spend most of its token budget on output.
Practical per-call estimates for a typical task (4,000 input tokens + 500 output tokens):
- Haiku 4.5: ~$0.0065
- Sonnet 4.6: ~$0.019
- Opus 4.7: ~$0.032
Scale that to 100,000 calls/month and the difference between Haiku and Opus is roughly $650 vs $3,200 — for identical task volume.
Claude Code Subscription Cost
Claude Code, Anthropic’s terminal-based coding agent, is bundled with Pro and above — there is no standalone free tier for it. What varies between plans is usage depth:
- Pro ($20/month): Claude Code access included, but complex multi-file or long-horizon tasks can exhaust the usage ceiling quickly.
- Max 5x ($100/month): 5x Pro’s usage, suited for active daily coding sessions.
- Max 20x ($200/month): For developers running Claude Code as their primary IDE assistant throughout the workday.
- Team Premium ($125/seat/month): Claude Code for teams with shared workspace and admin controls.
One clarification worth making: the subscription plan controls how much you can use Claude Code through the interface. If you build your own coding agent on top of the Anthropic API, that cost is separate and billed per token.
How API Token Costs Add Up
The token math gets non-obvious fast, especially with context-heavy tasks.
Context matters more than completion length. If you pass 100K tokens of context per call and generate only 2K tokens back, you’re spending $0.50 on Opus 4.7 input alone, per call. Developers building document Q&A, code review, or long-context summarization tools need to treat input cost as the primary variable, not output.
Agentic loops multiply tokens. A single Claude Code session that reads a codebase, revises files, and runs verification steps can generate 50K–200K tokens across the full conversation. Plan accordingly.
Model choice is the biggest cost lever. Before committing a workload to Opus 4.7, test whether Haiku 4.5 or Sonnet 4.6 achieves acceptable quality. Many classification, extraction, and summarization tasks run fine on Haiku at 80% lower cost per token.
Batch and Caching Discounts
Anthropic offers two structured cost reduction paths:
Prompt Caching
For repeated context (system prompts, reference documents, codebase snapshots), prompt caching avoids re-charging the full input price on cached tokens. Cache read rates are significantly lower — roughly 90% savings on the cached portion for high-reuse prompts. Cache writes have a small surcharge; the economics work when the same context block appears across many calls.
Batch API
The Batch API processes requests asynchronously and returns results within 24 hours. Current discount: 50% off both input and output tokens on supported models. This is a direct fit for eval pipelines, bulk content generation, nightly data processing, and any workload where real-time latency is not required.
Combining both: a document analysis pipeline that reuses a large system prompt and processes requests in batches can reduce effective token cost by 70–80% compared to synchronous, non-cached calls.
Choosing the Right Plan or Model Tier
A few decision rules that actually hold:
If you use Claude for chat and light coding a few times a week: Pro at $20/month is usually enough. The annual rate brings it to ~$17/month.
If Claude Code is your primary coding environment: Start at Max 5x ($100/month). Drop to Pro first and monitor whether you hit limits within a week — if you do, upgrade.
If you’re building an API-based product: Do not pay for a subscription as a proxy for API costs. They are separate. Buy credits, use the Batch API for non-realtime jobs, and pick the cheapest model that passes quality bar on your specific task.
For teams: Team Standard covers most collaboration needs. Team Premium makes sense only if engineers need Claude Code at full usage depth — the $100/seat annual rate roughly matches an individual Max 5x plan.
For enterprise: Negotiate on volume, not tier. The headline per-seat rate is a starting point, not a ceiling.
Using Novita AI LLM API as a Cost Alternative
For developers building with open-source models or running cost-sensitive workloads, Novita AI LLM API provides an OpenAI-compatible endpoint for models like Llama, Qwen, DeepSeek, and GLM at substantially lower token costs. Input pricing starts at $0.02/million tokens for smaller models, and capable mid-size models like Qwen3 and DeepSeek V3 series run under $0.50/million tokens for input.
This matters in practice for Claude Code users: several open-source models can be used as Claude Code backends through the API compatibility layer, reducing per-session token cost significantly. For example, GLM-5 runs at $1.00/$3.20 per million tokens on Novita AI — a fraction of Opus 4.7’s rate — and has been specifically tested for agentic coding workflows. See how to use GLM-5 in Claude Code for a setup guide.
For teams running multi-step agents, Novita Agent Sandbox provides managed execution environments for browser automation, code execution, and file operations — the infrastructure layer that API calls alone don’t cover. For pricing considerations on agent sandbox infrastructure specifically, see AI agent sandbox pricing models.
If budget optimization is the primary driver, see the full top 10 cheapest LLM APIs in 2026 for a ranked comparison across providers.
FAQ
How much does Claude cost per month?
Claude costs $0 (Free), $20 (Pro), $100 (Max 5x), or $200 (Max 20x) per month for individual plans. Team plans start at $25/seat (Standard) or $125/seat (Premium). Enterprise pricing is custom.
What is Claude’s API pricing?
As of July 2026: Haiku 4.5 at $1/$5 per million tokens (input/output), Sonnet 4.6 at $3/$15, and Opus 4.7 at $5/$25. All rates are pay-as-you-go with no monthly minimum.
How much does Claude Pro cost?
$20/month billed monthly, or approximately $17/month on an annual plan.
What is Claude Max pricing?
Max 5x is $100/month and Max 20x is $200/month. Both are individual plans for heavy Claude usage including Claude Code. No annual billing option is currently listed for Max tiers.
How much does Claude Code cost?
Claude Code is included in Pro ($20/month) and above. There is no free tier. Team Premium ($125/seat/month) is the recommended tier for engineering teams needing full Claude Code access.
What is Anthropic Claude Opus 4 pricing on the API?
Claude Opus 4.7 (the current Opus 4 release) is $5.00/MTok input and $25.00/MTok output. Prompt caching and Batch API discounts apply.
What is Claude Sonnet 4 pricing?
Claude Sonnet 4.6 is $3.00/MTok input and $15.00/MTok output.
Does Anthropic offer volume discounts?
Yes, through two mechanisms: the Batch API (50% off input and output for async processing) and prompt caching (up to 90% savings on repeated context blocks). Enterprise contracts may also include negotiated rate caps.
What are Anthropic API costs at scale?
At 10 million calls/month with 4K input + 500 output tokens per call: Haiku 4.5 totals ~$65,000/month before discounts; Sonnet 4.6 ~$195,000; Opus 4.7 ~$320,000. Batch API discounts cut these roughly in half for non-realtime workloads.
