How Anthropic structures enterprise pricing for Claude in 2026: usage tiers, committed-spend discounts, dedicated capacity, SLAs, and what to negotiate.
The list pricing on anthropic.com — $15/$75 per million tokens for Opus 4.7, $3/$15 for Sonnet 4.6, $1/$5 for Haiku 4.5 — is the on-demand consumption rate that most teams pay. Above roughly $30k–$50k per month in spend, Anthropic Sales will open a conversation about enterprise terms. This page describes how that pricing typically works in 2026 based on public statements and the Anthropic console's documented tiers, so you know what to ask for before getting on a call.
The four usage tiers (self-serve)
Even before you talk to Sales, every Anthropic account moves through automated rate-limit tiers based on payment history and successful API usage. Tier upgrades unlock higher RPM and TPM ceilings — they do not change the per-token price.
Tier
Trigger
Sonnet 4.6 limits (typical)
1
First payment of any amount
50 RPM · 40k input TPM · 8k output TPM
2
$40 cumulative spend + 7 days
1,000 RPM · 80k input TPM · 16k output TPM
3
$200 cumulative spend + 7 days
2,000 RPM · 160k input TPM · 32k output TPM
4
$400 cumulative spend + 14 days
4,000 RPM · 400k input TPM · 80k output TPM
The exact numbers vary by model and shift over time — check the Anthropic console for live limits. Tier 4 is the ceiling reachable without sales contact. Beyond that, you need Custom.
Custom / Enterprise tier (sales-led)
Once you exceed Tier 4 limits or want contractual SLAs, the conversation moves to a custom agreement. The typical components in 2026:
Committed spend. A monthly or annual minimum (e.g. $100k/mo) in exchange for a discount off list — typically 5–25% depending on commit size and duration.
Dedicated capacity (Provisioned Throughput). A reserved RPM/TPM allocation that doesn't compete with the public pool. Priced as a monthly fee per unit of capacity rather than per-token. Useful if you need predictable latency under load.
Latency SLA. Contractual p95/p99 latency targets, with service credits on miss.
Availability SLA. Typically 99.5% or 99.9% monthly uptime per model.
Dedicated support and a shared Slack/Teams channel. Named technical contact, escalation paths, roadmap previews.
Custom data retention. Zero data retention (ZDR) is available on most enterprise contracts — Anthropic processes the request and discards inputs/outputs immediately, no 30-day retention window.
HIPAA BAA, EU data residency, IL4/IL5 (AWS GovCloud via Bedrock). Compliance riders depending on industry.
Pricing math on a $1M/year commit
Suppose you run 100M Sonnet 4.6 input tokens and 25M output tokens per day. At list:
Input: 100M × $3/M = $300/day
Output: 25M × $15/M = $375/day
Daily: $675 · Monthly: ~$20.5k · Annual: ~$246k
That's well above Tier 4 but below the typical commit threshold. To reach $1M/year ($83k/mo) you'd need ~4× this volume — at which point a 10–15% committed discount is realistic, saving $100k–$150k annually before any caching/batch optimisation.
What to negotiate
The discount percentage off list — primary lever. Anchor on your forecasted 12-month spend, not last month's.
Whether caching/batch discounts stack with the commit discount — they should, but get it in writing.
Overage rate — what happens if you exceed the commit. Should be list, not punitive.
Ramp schedule — start at a lower commit and step up, rather than a flat 12-month minimum.
Cancellation / down-step rights — annual commits should allow renegotiation if your usage drops materially.
Self-serve teams should still optimise list-price spend before talking to Sales — caching alone can cut a $100k bill to $30k, which is more leverage than a 15% commit discount. Walk through the cost optimization checklist first.
Frequently asked questions
What is the cheapest way to access Claude at enterprise volume?
List price minus 10–25% via a committed-spend contract once you're consistently over ~$50k/mo. But the largest savings come from operational optimization (prompt caching, model routing, batch) before commit discounts — most teams cut spend 60–80% with those three alone, dwarfing any negotiated rate.
Does Anthropic offer volume discounts on Claude API usage?
Yes, through committed-spend agreements negotiated by Sales. Typical discounts are 5–25% off the list per-token rate, scaling with commit size and contract duration. There is no public published volume table — discounts are individually negotiated above ~$30–50k/mo spend.
Can I get a Claude API SLA for production?
Latency and availability SLAs are available on enterprise contracts (typically 99.5% or 99.9% monthly availability per model, with p95/p99 latency commitments). Self-serve tier 1–4 accounts do not have a contractual SLA, only the public best-effort uptime page.
What is zero data retention on the Claude API?
ZDR is an enterprise option where Anthropic processes API requests and discards inputs and outputs immediately rather than retaining them for the standard 30 days. Required for HIPAA workloads and regulated industries. Available on custom contracts; not available on self-serve tiers.