What Claude Enterprise actually costs in 2026
Anthropic shifted to usage-based billing in November 2025. Here's what the new bill looks like.
Claude Enterprise is $20 per user per month plus whatever your team burns in API tokens. That's the model as of November 2025, when Anthropic killed bundled tokens and moved to pure usage-based billing. The $60/seat figure that tech press quoted in early 2025 refers to a plan that no longer exists.
The plan. $20 per seat on annual terms. 20-seat minimum for self-serve, 50-seat minimum for sales-assisted contracts with HIPAA-ready terms and product bundling. A 20-person team on self-serve pays $4,800/year in seat fees. ChatGPT Enterprise, by comparison, doesn't publish a list price. Secondary reporting puts actual seat rates in the $40 to $75 range depending on contract scale; minimums vary.
Usage is billed separately at standard API rates.
Per million tokens:
- Haiku 4.5: $1 in / $5 out
- Sonnet 4.6: $3 in / $15 out
- Opus 4.7: $5 in / $25 out
Two footnotes matter. Prompt caching cuts cached input reads by 90%; batch API cuts both input and output in half. Opus 4.7 also ships with a new tokenizer that consumes up to 35% more tokens for the same text. If you budgeted Opus 4.7 at Opus 4 token counts, you'll be 35% short.
How much usage in practice depends on workload pattern.
Chat-style deployments are cheap. A power user burns $10 to $30 in tokens a month on Sonnet 4.6. For a 50-seat rollout: $500 to $1,500/month in tokens on top of $1,000 in seats.
Agent deployments are categorically different. Anthropic's engineering team published the multiplier for their own multi-agent research system: "about 15× more tokens than chat interactions." At that ratio, a 50-seat Sonnet rollout runs $7,500 to $22,500/month in tokens.
Here's my own data point at the other end of the scale. I operate an agent system in production that runs on Claude. A dozen-plus scheduled agents handle content ingestion, workflow automation, and scheduled reporting, mixing Haiku for categorization, Sonnet for reasoning, and Opus sparingly for long-context work. Steady-state token spend runs under $20/month. Build-out was a few hundred dollars total, including a one-time backfill of two years of content history against Sonnet and Haiku. Less than several of the SaaS tools the system replaced.
Claude Code is the most public data point in between. Faros reports average usage of ~$6 per developer per day on Sonnet, with 90% of users under $12/day. The figure is directional, not Anthropic-published. Apply that to 20 devs × 22 working days: ~$2,640/month in tokens, on top of $400 in seats.
The question to answer before signing isn't "what's the per-seat cost." It's "what's our ratio of agent work to chat, and what's our token burn profile per workflow." Most teams scope by headcount. Then Q2 arrives.