Anthropic's Claude Partner Network has three Services Track tiers. Select requires at least 10 active certified individuals, 2 joint customers deployed in production in the trailing 12 months, and 1 public customer story. Preferred requires at least 100 certified individuals, 15 deployed joint customers, and 3 public stories. Global Premier requires at least 1,000 certified individuals, 100 deployed customers across three or more regions, 15 public stories, and a joint business plan with named executive sponsors.
Those are the published thresholds. The more useful question for anyone about to sign a statement of work is what a tier badge actually proves — and what it quietly leaves out. This guide covers both: each tier in detail, how firms gain and lose one, how to read a tier as a buyer, and the evaluation checklist we would hand a client choosing between three shortlisted firms.
What is the Claude Partner Network Services Track?
Anthropic launched the Claude Partner Network on 12 March 2026, backed by a stated $100 million committed for 2026 in partner training, technical support, and shared marketing. Membership is free. In June 2026 it added the piece that matters to buyers: a formal Services Track with published tier requirements, plus the Claude Partner Hub, which works as both a partner-facing dashboard and a customer-facing directory.
The scale of the response tells you why the tiering exists. Anthropic reports that more than 40,000 firms have applied to join and more than 10,000 consultants have earned a Claude certification. When forty thousand companies can call themselves "Claude partners," the label stops carrying information. The Services Track replaces a binary badge with thresholds you can check, and the Hub's public directory shows each partner's tier, certified team, customer deployments, and public references.
One detail is worth pulling out of the fine print, because it is the most anti-gaming rule in the program: a certified individual counts only if they hold a current Anthropic certification and have used Claude in the past 90 days. Certifications collected and shelved do not count.
What do the three Services Track tiers require?
Every tier is measured on the same three axes — people, deployments, and public evidence. The bar moves by roughly an order of magnitude at each step.
Select: the entry tier
Select is the first rung with a published bar: 10 active certified individuals, 2 joint customers deployed in production in the trailing 12 months, and 1 public customer story.
Read what that actually asks for. Ten people is a team, not a department — reachable by a boutique that has genuinely committed to Claude, out of reach for a firm that sent two people on a course. Two production deployments separates talkers from shippers, and the trailing-12-month window means they have to be recent. One public story means at least one customer agreed to be named.
For a buyer, Select is a floor, not a ceiling: the firm has shipped Claude work for at least two customers in the last year and keeps a bench of at least ten current practitioners. Whether any of them have built your workflow is a separate question the tier does not answer.
Preferred: a systematized practice
Preferred asks for 100 active certified individuals, 15 deployed joint customers, and 3 public stories — a tenfold jump on people and roughly sevenfold on deployments.
The interesting threshold is 100 certified individuals, because keeping 100 people currently certified and actively using Claude is an operational commitment, not a marketing decision. It implies internal enablement, a training budget, and someone whose job is tracking certification currency. Fifteen production deployments implies repeatable delivery: patterns, reusable scaffolding, a sales motion that has found its shape.
This is the tier where you should expect process artifacts — ask to see the delivery method, the QA gate, and the handover package. A firm that has done this fifteen times and cannot show a standard approach has run fifteen bespoke projects, a different risk profile.
Global Premier: scale across regions
Global Premier requires 1,000 active certified individuals, 100 deployed joint customers across three or more regions, 15 public customer stories, and a joint business plan with Anthropic backed by named executive sponsors.
The multi-region clause is the one to notice. It is not a headcount test; it tests whether the firm can staff and support delivery in more than one part of the world. If your rollout spans jurisdictions with different data residency rules, that clause is doing real work for you. If your project is one workflow in one country, it buys you nothing.
The joint business plan is worth reading for what it is: alignment between two vendors on go-to-market, not a commitment about your account. Anthropic and the firm have a shared plan. That does not mean your escalation gets a partner's phone number.
How firms gain and lose a tier
Tier standing is recalculated on a fixed cadence rather than continuously. Anthropic published promotions on 1 January and 1 July, with an additional review on 1 October 2026 in this first year. Demotions are reviewed annually on 31 December, with 90 days' notice. Partners can watch their own standing against the thresholds in the Partner Hub, refreshed daily.
Three consequences follow for buyers. A badge reflects a trailing snapshot, not today — ask when the firm was last reviewed and what has changed since. A firm that grew fast last quarter may be stronger than its tier suggests, so a recent Select can outperform a stale Preferred. And because deployments count over a trailing 12 months while certifications drop out of the count after 90 days of non-use, a tier measures current activity that has to be re-earned. A firm coasting on work it did two years ago drifts down.
What the tiers measure — and what they don't
The tiers are a real improvement over self-declared "certified partner" claims because they are outcome-based. Certified people who used the product recently, customers actually live in production, and named references who agreed to go on the record are all hard to fake at scale.
What a tier does not encode is fit. Here is the sharpest way to put it: the thresholds measure committed Claude capacity, not firm size, and neither one is the same as your delivery team. A 5,000-person systems integrator with 100 certified practitioners and a 150-person specialist with 100 certified practitioners land in the same tier — but in the first case that is 2% of the firm and in the second it is most of it. The tier cannot tell those apart, and for your project the concentration may matter more than the total.
Read the requirements literally and the rest of the ambiguity is obvious. "Active certified individuals" is a firm-wide headcount, not a promise about who staffs your account. "Customers deployed in production" counts deployments, not their scale or durability. "Public customer stories" counts published references, a marketing artifact as much as a quality one. Every metric is a useful proxy. All of them are still proxies.
What partners get from the network
Understanding the partner's side helps you read their incentives. Getting started is free; firms gain Anthropic Partner Academy access including certification exams, with discounted exam rates and priority access to new certifications for tiered partners. At launch Anthropic also described a partner portal with training and sales playbooks, the Claude Certified Architect, Foundations certification, access to Applied AI engineers, and co-marketing support; Accenture, Deloitte, Cognizant, and Infosys were quoted in the announcement. So a tier is worth real money to the firm holding it, through directory placement and co-marketing. That is not a reason to distrust it — it does explain why the badge leads the first slide of every deck.
So which tier should you hire?
Match the tier to the shape of the work, not to your ambition. For a scoped pilot on one or two workflows, a Select-tier specialist with directly relevant production references will usually move faster and cost less than a large firm's junior bench. For a multi-region rollout with formal change management, procurement scrutiny, and a five-year horizon, the depth and continuity of a Preferred or Global Premier firm earns its premium.
Tier gets a firm onto your shortlist. These eight questions pick between them — score every firm on all eight before you compare prices.
- Named delivery team, not firm-wide stats. Ask who will be on your account, their certifications, and what they shipped in the last 90 days. A firm-wide practitioner count is not an answer.
- Production references in your shape of problem. Not "in your industry" as a vague claim — a workflow with comparable data messiness, integration surface, and volume. Ask to speak to one.
- Definition of done. Does the engagement end at a demo, a pilot, or a system with an owner, monitoring, and an exception path? This single question separates most vendors.
- Governance before scale. Access boundaries, permissions, approval rules, and data policy belong at pilot stage, not retrofitted after rollout.
- Cost model literacy. A partner who cannot explain what drives your run-rate — model routing, context size, retries, tool-call frequency — will hand you a bill you cannot forecast.
- Measurement defined upfront. Agree the metric before build starts: cycle time, hours reclaimed, error rate, cost per transaction. "Improved productivity" is not a metric.
- Handover and independence. Can your team operate and debug the system without the partner? Ask what documentation and access you get at close.
- What happens in month seven. Who owns the automation when a source system changes its API? Get the support model in writing.
Most weak proposals fail on items three and eight. Proof-of-concept work is the easiest thing to sell and the least valuable thing to own.
How ATI approaches Claude implementation
Straight answer, since this is a piece about reading partner credentials: ATI does not hold or claim a Claude Partner Network tier, and we are not asking you to buy on a badge. We would rather be judged on the eight questions above, so here is how we answer them.
Our Claude for business approach starts with one pilot before any wider rollout, organised into three workstreams — value, governance, and enablement — so security, leadership, and delivery teams align on controls and access boundaries before scale, not after. Delivery runs in three phases: strategy and readiness (auditing workflows, tools, and data to find where AI actually fits), applied build (agents and integrations piloted behind a human-in-the-loop QA gate), then production delivery with defined ownership, observability, and exception handling. Typical first deployment runs 6–12 weeks.
On outcomes, the numbers we publish are three specific workflows out of 25+ in production — named rather than averaged, and not a benchmark to expect for your process: shipment intake with manual work cut 95% and zero monthly entry errors, an ERP assistant that cut support cost 90% and replaced $10k+/month of repetitive support, and email-to-CRM that made project setup 80% faster. Across the practice, 98% long-term retention. We are measured on hours reclaimed, error rate, and cost — not dashboards.
On cost, we treat spend as three layers rather than one number — usage (model calls, context size, retries), runtime (infrastructure, orchestration, observability), and ops (human review, exception handling). The full breakdown is in our AI agent cost guide, and teams standardising on Claude Code across engineering can see the operating model in Claude Code for companies.
Three answers should end a conversation: a tier badge offered as the answer to a capability question, a tier claim with no review date, and a case study with a percentage but no workflow, baseline, or timeframe.
The pattern worth avoiding is buying tier as insurance. It transfers a defensible-sounding decision onto a badge, and badges do not attend your standups. The firms that deliver are the ones that can describe, before you sign, exactly what will be in production, who will own it, and how you will both know it worked.
Frequently asked questions
What are the Anthropic partner tiers?
Three, in the Claude Partner Network's Services Track: Select, Preferred, and Global Premier. They are ranked on the same three axes — active certified individuals, joint customers deployed in production, and public customer stories — with Global Premier adding a multi-region requirement and a joint business plan with named executive sponsors.
What is an Anthropic Select partner?
The entry tier of the Services Track. A Select partner has at least 10 active certified individuals, at least 2 joint customers deployed in production in the trailing 12 months, and at least 1 public customer story. It confirms recent, real delivery — not the size of the firm or its fit with your use case.
What does Global Premier partner status mean?
The top tier: at least 1,000 active certified individuals, at least 100 deployed joint customers across three or more regions, at least 15 public customer stories, and a joint business plan with Anthropic backed by named executive sponsors. It signals scale and multi-region delivery capacity. For a single-country pilot, a smaller firm with directly relevant references may deliver faster and cost less.
What counts as a certified individual?
Someone at the firm who holds a current Anthropic certification and has used Claude within the past 90 days. The recency rule is deliberate — it prevents firms from counting certifications earned once and never used.
How often do Anthropic partner tiers change?
Anthropic published promotions on 1 January and 1 July, with an additional review on 1 October 2026 in the program's first year, and an annual demotion review on 31 December with 90 days' notice. Partners track their own standing in the Claude Partner Hub, refreshed daily. Any badge you see reflects the last review, so ask a prospective partner when they were last assessed.
Do I need a partner at all, or can we implement Claude ourselves?
Plenty of teams run their own pilots successfully. Outside help pays for itself at the move from a working prototype to a system with an owner, monitoring, exception handling, and a forecastable run-rate — the phase where most internal AI projects stall. To pressure-test where your first workflow sits on that path, book a strategy call.
Tier requirements, review dates, and program benefits in this guide are quoted from Anthropic's own Claude Partner Network announcements and were verified on 4 August 2026. Anthropic sets these terms and can change them — verify current details on Anthropic's partner pages before making a procurement decision.