Last updated: August 3, 2026
Material correction — August 3, 2026
This page was originally published on April 4 and updated on April 17 under the title “Claude Pro and Max Just Lost Their Most Powerful Feature.” The original version described Anthropic’s April change as a universal cutoff of OpenClaw and all third-party tools from Pro and Max subscription limits. That framing was too broad and should not be relied on.
Our subsequent primary-source audit found that API-key access, native Claude Code authentication, local claude -p, setup-token custody, paid-plan Usage Credits, and Anthropic’s treatment of certain third-party tools must be analyzed as separate routes. We have therefore replaced the article in full.
As of August 3, 2026, FSR had not received a response to its route-specific Anthropic Support request. Any later response will be appended with its date, scope and evidentiary status rather than silently folded into the article.
Claude Pro and Max are consumer subscriptions for Claude and Claude Code. Anthropic documents several non-interactive and third-party use cases, and it does not market either plan as a portable entitlement for every external client. Evaluate a route by identifying the active credential, who stores it, which process sends the request, which balance pays, which terms apply, and whether prior approval is required.
Verdict: upgrading from Pro to Max buys more documented capacity and priority benefits. The reviewed public record does not show that it expands third-party permission, and it does not resolve OpenClaw’s specific status as of August 3, 2026.
What happened
In April 2026 the billing treatment of third-party harness usage changed and was widely reported as a total cutoff. Anthropic later proposed a separate monthly Agent SDK credit, then paused that change on June 15 before it took effect. Its current Agent SDK page says Agent SDK, claude -p and third-party app usage still draw subscription limits for now.
Who this affects
- Subscribers running agents against their own Claude login
- Anyone weighing Pro against Max 5x or Max 20x
- Anyone who found an API charge next to an active subscription
- Developers choosing between subscription auth and an API key
Who this does not affect
- Anyone using Claude, Claude Code or Cowork through Anthropic’s own apps only
- Anyone already on a separately funded Console API key
- Team and Enterprise administrators, whose contract layer differs
Where this analysis stops
- It does not rule on whether any specific tool is permitted
- It offers no legal, contract or compliance advice
- It reports no benchmark, latency or cost-per-task figure
- It covers individual plans, not procurement or data-retention review
Make the decision in this order
| If your goal is | Start here |
|---|---|
| More capacity in Claude or subscription-authenticated Claude Code | Compare Pro, Max 5x and Max 20x |
| A documented authentication, billing and contract path for an integration | Use a separately funded Anthropic API route |
| Finding the source of an unexpected API charge | Check Claude Code’s active authentication method and the credential precedence chain |
| Using OpenClaw through a subscription credential | Treat permission, custody and billing as three separate questions |
| Removing OpenClaw-specific policy uncertainty | Obtain written route-specific confirmation, or use the documented API route |
Key facts
| Pro | $20 per month (US); at least five times free usage per session; Claude Code and Cowork included |
| Max | Max 5x $100 and Max 20x $200 per month on the web; 5x or 20x Pro usage per session; priority access |
| API access | Not included in any paid Claude subscription; Console is a separate product with separate payment |
| Billing buckets | Three: included plan limits, paid-plan Usage Credits, separately funded Console API billing |
| Routes audited | Ten access routes plus one billing overlay |
| Support request | Route-specific question sent to Anthropic; unanswered as of 3 August 2026 |
On this briefing
Contents
Route before tier
Start with the route. API keys, native subscription OAuth, claude -p, Agent SDK use and setup-tokens do not produce the same answer.
The phrase “third-party access” hides eight decisions: actor, credential origin, custody, runtime authority, request sender, balance, terms and approval. Change one and the documented answer can change with it. Keep capability, billing and permission in separate records. Source code establishes capability. Billing documentation identifies a meter. Anthropic policy states Anthropic’s position, and only Anthropic’s policy does that.
Direction matters as much as the variables. This briefing covers non-Anthropic software sending requests to Claude. Claude reaching outward through connectors or MCP is the opposite traffic direction and a different set of documents.
Was Claude Pro or Max universally cut off from OpenClaw? The audited Anthropic pages describe several distinct routes rather than a single cutoff, and the current Agent SDK page describes third-party app usage as still drawing subscription limits for now. Those same pages publish no approval for OpenClaw. What Anthropic does publish is a default restriction on third-party developers offering Claude.ai login or subscription rate limits, a discretionary allowance for certain third-party tools used by subscribers with usage credits enabled, and a reserved right to bill that use to Usage Credits. None of the three names OpenClaw.
Sources: Anthropic Help Center, 16 June 2026 · Anthropic Help Center, 19 May 2026 · Anthropic, Claude Code Docs, Agent SDK overview, accessed 3 August 2026
What Pro and Max actually change
Both plans buy capacity and priority on Anthropic’s own surfaces, and both exclude the API.
Anthropic lists Pro at $20 per month in the United States, with at least five times the free service’s usage per session, priority access during high-traffic periods, early access to new features, Claude Code and Cowork. Max is offered at $100 for Max 5x and $200 for Max 20x on the web, with five or twenty times Pro’s usage per session, priority access to new models and features, and the same Claude Code and Cowork inclusion. The documented differences are capacity, priority access and weekly-limit structure. No reviewed source states a permission difference between the tiers.
When a Pro or Max subscription is the active credential, Claude and Claude Code share the plan’s usage limits, and Anthropic states that IDE sessions count against the same pool. Supported editors include VS Code, Cursor and other VS Code forks, and JetBrains editors such as IntelliJ and PyCharm. A first-party Claude Code client running inside a third-party editor is a different arrangement from a third-party client holding your credential, and the documentation treats them differently.
Neither plan includes the API. The Pro page states that the plan does not include API usage through the Claude Console, and a separate article states that a paid Claude subscription doesn’t include access to the Claude API or Console. That rules out one common buying mistake: upgrading to Max because an external tool appears to need a broader entitlement. Max matters after a route is eligible and metered against subscription limits.
Sources: Anthropic Help Center, 10 June 2026 · Anthropic Help Center, Max plan, accessed 3 August 2026 · Anthropic Help Center, 11 June 2026 · Anthropic Help Center, 16 March 2026
Five route families
Five families cover most buyer situations, and the differences between them are documented in different places.
1. API-key access. An Anthropic API key belongs to the Console and API product rather than to a Pro or Max subscription. Requests are billed separately at the applicable API rates under Anthropic’s Commercial Terms, and the subscription tier neither funds nor enlarges the route. Commercial Terms A.1 grants permission to use the services including to power products a customer makes available to its own customers and end users. This is the clearest public route for a developer building an integration or a product, where “clearest” refers to its documented authentication, billing and contract structure rather than to uptime or latency.
2. Native Claude Code subscription authentication. A subscriber signs in to Claude Code with the subscription account, and Claude and Claude Code then share the plan’s limits. An installed credential can change that result. Claude Code follows a published precedence chain, and a signed-in Claude apps gateway session, configured cloud-provider credentials, an ANTHROPIC_AUTH_TOKEN or an effective ANTHROPIC_API_KEY can each take precedence over stored subscription OAuth. Holding a subscription therefore does not prove which balance paid for a specific request.
3. A local controller invoking claude -p. Anthropic’s current Agent SDK page states that Agent SDK usage, the claude -p command and third-party app usage still draw from subscription usage limits, framed as the state For now, nothing has changed after a proposed billing change was paused on June 15, 2026. That identifies a current meter in the Agent SDK context. It does not answer the permission question for every controller that can spawn the command, and the reviewed pages do not map a subscriber’s own controller against the restriction that applies to third-party developers.
4. A setup-token held by another application. Anthropic documents claude setup-token as a way to create a one-year OAuth token for CI pipelines and scripts where interactive login is unavailable. The command prints the token, saves no copy, limits it to model requests, and requires a paid plan. Pasting that bearer token into another application changes credential custody, which is a different question from permission and from metering.
5. A product offering Claude.ai login or subscription rate limits. Anthropic’s Agent SDK documentation directs product builders to API-key authentication and states that it does not allow third party developers to offer claude.ai login or rate limits for their products unless previously approved. Its login guidance separately acknowledges that Anthropic may, at its discretion, allow paid subscribers with Usage Credits enabled to use certain third-party tools. The reviewed pages publish no approved-tool list, approval criteria, mapped billing treatment or revocation terms, and they do not name OpenClaw.
Among the factors these pages name for third-party tool use are subscriber status, whether Usage Credits are enabled, whether the tool represents its identity honestly, and Anthropic’s discretion. That list is what the pages state, not a complete or sufficient eligibility test.
Sources: Anthropic, Commercial Terms of Service, effective 17 June 2025 · Anthropic, Claude Code Docs, Authentication, accessed 3 August 2026 · Anthropic Help Center, 16 June 2026 · Anthropic, Claude Code Docs, Agent SDK overview, accessed 3 August 2026 · Anthropic Help Center, 19 May 2026
Technical appendix: ten routes and one billing overlay
| Route | Credential custody | Request sender | Documented permission | Which balance may pay | Terms layer | Pro/Max difference | Evidence |
|---|---|---|---|---|---|---|---|
| 1. API key, personal use | Developer’s own environment or secret store | Developer’s own process | Documented route | Separately funded Console API billing | Commercial Terms | Tier does not participate | A05, A07, A12 |
| 2. API key powering a product for the developer’s customers | Vendor infrastructure | Vendor infrastructure | Documented route under Commercial Terms A.1 | Separately funded Console API billing | Commercial Terms | Tier does not participate | A05, A07 |
| 3. Native Claude Code subscription OAuth | macOS Keychain, Linux credentials file at mode 0600, or Windows user profile | Local Claude Code process | Documented route; OAuth described as intended for plan purchasers and native Anthropic applications | Included plan limits, then paid-plan Usage Credits if enabled | Consumer Terms govern the subscriber relationship | Capacity, priority access, weekly-limit structure | A03, A04, A13 |
4. Generic third-party controller invoking local claude -p | Claude Code’s own store, unless a higher-precedence credential is present | Local Claude Code process | Unresolved in the audited public sources for this actor | Currently described as included plan limits in the Agent SDK context; an effective API key selects Console API billing | Consumer Terms for the subscriber; Agent SDK material points to Commercial Terms; interaction not mapped | No source states a permission difference | A02, A03, A04 |
| 5. OpenClaw native-login import plus passthrough | Imported material may be persisted in OpenClaw’s per-agent SQLite store or held runtime-only; the passthrough branch forwards nothing | Spawned Claude Code process | Agent SDK context describes third-party app usage drawing plan limits; the OpenClaw-specific mapping is not published | Not published for this arrangement | Consumer Terms for the subscriber; OpenClaw’s own terms separately | No source states a permission difference | A02, B05–B09 |
| 6. setup-token stored by OpenClaw, direct runtime | User pastes the token; OpenClaw registers it in its own auth profile store | OpenClaw’s own model runtime | Official mechanism for the subscriber’s own CI and scripts; not established for this arrangement | Not published for this arrangement | Not conclusively established in the audited sources | Token requires a paid plan; no tier difference documented | A04, B10 |
| 7. setup-token forwarded to a configured Claude CLI backend | OpenClaw’s store, then passed to the spawned process over a file descriptor | Spawned Claude Code process | Not established for this arrangement | Not published for this arrangement | Not conclusively established in the audited sources | No tier difference documented | A04, B11 |
| 8. Agent SDK with API-key authentication | Process environment | Developer’s own process | Documented route; the Quickstart directs developers here | Separately funded Console API billing | Commercial Terms | Tier does not participate | A05 |
| 9. Agent SDK with subscription authentication | Process environment | Subscriber-developer’s own process | Partly documented; the help page describes the meter, the SDK page directs product builders to API keys | Currently described as included plan limits | Consumer Terms govern the subscriber; the SDK page points to Commercial Terms; the layering is not mapped | A tier-scaled credit was proposed and paused before implementation; no current rule is in effect | A02, A03, A05 |
| 10. A product offering Claude.ai login or subscription rate limits | Vendor holds each user’s subscription credential | Vendor infrastructure | Prohibited by default unless previously approved | No default included-plan entitlement is documented for an unapproved product; approved exceptions are acknowledged but their meter is not mapped | Commercial Terms for the vendor; Consumer Terms for each user | The restriction is not tier-specific | A03, A05 |
| Billing overlay — paid-plan Usage Credits and Usage Bundles | The Claude consumer account | Whichever process the underlying route uses | Official mechanism; opt-in and prepaid | Paid-plan Usage Credits at standard API rates; Bundles are discounted prepayment into the same pool | Consumer Terms | No tier-specific credit rule is currently in effect | A09, A15 |
Evidence key for the table above
All sources opened 3 August 2026. The full list, including sources not cited in this table, appears at the end of this page.
Rows 6 and 7 reuse the same token but change the request sender. Rows 4 and 5 both spawn Claude Code but differ in credential custody. The final row is a billing overlay rather than an access route, because it has no distinct actor, credential or sender of its own.
Sources: Anthropic, Claude Code Docs, Legal and compliance, accessed 3 August 2026 · Anthropic, Claude Code Docs, Authentication, accessed 3 August 2026 · Anthropic, Consumer Terms of Service, effective 8 October 2025 · Anthropic Help Center, 18 May 2026
Credential custody inside OpenClaw
OpenClaw’s source code at a fixed snapshot establishes what the project implements. It does not establish a successful runtime request, Anthropic’s billing treatment, or Anthropic’s permission.
At snapshot e812630f, OpenClaw can read compatible credential material from the host Claude login. The function that does so is documented in the repository as reading in Claude Code’s own precedence order, which means an API-key helper first, then the macOS Keychain, then the local credentials file. A second module converts that material into auth-profile entries carrying access, refresh and expiry fields. A third synchronizes those profiles, marks each as persisted or runtime-only according to the provider configuration, and refuses the import when the identity does not match. Persistence is conditional. Storage, when it happens, is a per-agent SQLite table named auth_profile_store.
In OpenClaw’s native-login passthrough branch, the imported credential is not forwarded to the subprocess. The code verifies that the live host login matches the selected identity, fails the run if it does not, then clears the credential before spawning. The repository comment gives the reason: a forwarded snapshot goes stale and blocks the subprocess from refreshing itself. The spawned claude -p process authenticates from Claude Code’s own store instead.
The setup-token path works differently. Anthropic mints a one-year OAuth token, prints it, saves no copy and limits it to model requests. When a user pastes that token into OpenClaw, the project’s non-interactive path writes it into its auth profile store as a token credential under its Anthropic provider. From there, OpenClaw implements two branches. Its own runtime can send the request, or a separately configured Claude CLI backend can supply the token to a spawned claude -p process over a file descriptor. The same file states that imported native logins are never forwarded, because those runs go through identity-verified passthrough.
One detail deserves a reader’s caution. OpenClaw’s CLI backend descriptor carries a code comment describing how Anthropic bills subscription-OAuth traffic and asserting that opted-in embedded runs execute on plan limits. That is an assertion inside a third-party repository. No Anthropic-controlled page in the audited set corroborates it, and it is recorded here as OpenClaw’s stated assumption rather than as Anthropic’s billing treatment.
Sources: OpenClaw, cli-credentials.ts at e812630f, accessed 3 August 2026 · OpenClaw, cli-migration.ts at e812630f · OpenClaw, external-cli-sync.ts at e812630f · OpenClaw, sqlite.ts at e812630f · OpenClaw, prepare.ts at e812630f · OpenClaw, register.runtime.ts at e812630f · OpenClaw, cli-backend.ts at e812630f · Anthropic, Claude Code Docs, Authentication, accessed 3 August 2026
Three billing buckets
Three Anthropic billing buckets matter to an individual subscriber, and Usage Bundles are not a fourth.
Included Pro or Max usage limits come with the plan and reset on a five-hour session window alongside weekly limits. Paid-plan Usage Credits are a consumer overflow layer that must be enabled and prepaid, are charged separately at standard API rates, and carry a monthly spend cap or an unlimited setting, optional auto-reload and a stated daily redemption ceiling of $2,000. Usage Bundles are discounted prepayment into that same pool, at 10, 20 and 30 percent for $50, $250 and $1,000, capped at $2,000 of discounted bundles per month for individual Pro and Max subscribers, and Anthropic states that bundle credits apply only after plan limits are exceeded. Separately funded Console API billing is the third bucket, attached to an API key and a different contract.
Which bucket pays depends on the effective credential, not the plan label. Anthropic’s documented precedence places cloud-provider credentials first when the relevant environment variable is set, then ANTHROPIC_AUTH_TOKEN, then ANTHROPIC_API_KEY, then an apiKeyHelper script, then CLAUDE_CODE_OAUTH_TOKEN, then subscription OAuth. A signed-in Claude apps gateway session sits outside that list as a provider selection and outranks the rest. If ANTHROPIC_API_KEY is the effective credential under that chain, Claude Code bills the API account attached to that key. Higher-precedence credentials do not all route to Anthropic’s Console.
Anthropic documents consent controls on the transitions it describes. The usage-credits page describes prior enablement, funding, a notification on approaching and reaching the limit, and a confirmation before continuing. For credential conflicts, Anthropic states that All transitions to API credit usage require explicit user consent, and describes a setup-time confirmation when a key is detected plus a post-login notification when both credentials are active. It also names the failure mode plainly: setting the variable in a shell configuration file and forgetting about it. What the public pages leave unmapped is how a remembered consent state carries into every headless configuration, including non-interactive runs where the documentation states the key is used whenever present.
One page widens the picture. Anthropic’s usage-bundles article describes the bundle balance as a single pool applying across Claude, Claude Desktop, Claude Mobile, Claude Code, Cowork and third-party products that use your Claude account, without naming which products qualify.
Sources: Anthropic Help Center, 18 May 2026 · Anthropic Help Center, 18 May 2026 · Anthropic, Claude Code Docs, Authentication, accessed 3 August 2026 · Anthropic Help Center, 5 May 2026 · Anthropic Help Center, 11 June 2026
Four unresolved public mappings
Four gaps sit in the public record. They are missing mappings rather than proven contradictions, and this briefing leaves them open.
1. Default meter versus discretionary re-metering. The June notice says Agent SDK, claude -p and third-party app usage still draw subscription limits for now. The login guidance reserves discretion to charge certain third-party tools to paid-plan Usage Credits instead. The credits and bundles pages describe a post-limit overflow layer. No page publishes the tool-by-tool trigger that joins the three.
2. Agent SDK contract layering. Consumer Terms govern the subscriber relationship for Free, Pro and Max. The Agent SDK page points to Commercial Terms for SDK use. The June help article describes SDK use drawing consumer subscription limits. How those layers interact for a personal project on a consumer plan is not published.
3. Automated access and third-party app metering. Consumer Terms restrict access through automated or non-human means except through an API key or where Anthropic explicitly permits it. The June help article describes third-party app usage drawing subscription limits for now. The route-by-route class mapping that reconciles the two is not published.
4. OpenClaw attribution. OpenClaw publishes an attribution to unattributed, non-public staff guidance. The reviewed Anthropic pages publish no matching OpenClaw-specific statement. Those are asymmetric source records, not proof about either side.
OpenClaw’s current documentation states that Anthropic staff told us this usage is allowed again and treats Claude CLI reuse and claude -p as sanctioned for its integration unless Anthropic publishes a new policy. That language entered the repository on April 6, 2026, in a commit that also removed the project’s earlier framing about separately billed extra usage. It carries no named person, no date and no linked statement.
OpenClaw’s source can establish what the project implements, and its documentation can establish what the project reports. Neither substitutes for Anthropic’s public position. FSR sent Anthropic a route-specific Support question covering actor, client, authentication, credential custody, execution, billing, approval, governing terms and effective date, and no response had arrived by the publication lock. Silence is not treated as an answer in either direction.
Sources: OpenClaw documentation, OAuth, accessed 3 August 2026 · OpenClaw, commit d378a50, 6 April 2026 · OpenClaw, commit 37301cb, 4 April 2026 · Anthropic Help Center, 19 May 2026 · Anthropic, Consumer Terms of Service, effective 8 October 2025
Risk, enforcement and recourse
Anthropic publishes an enforcement reservation and an appeal route, and neither predicts anything about a specific tool.
The Claude Code legal page states that advertised Pro and Max limits assume ordinary, individual usage of Claude Code and the Agent SDK and reserves the right to enforce restrictions without prior notice. No audited page quantifies that assumption or ties any runtime pattern to a determination, and no threshold should be inferred from the phrase.
Anthropic also publishes three named grounds for banning an account: repeated Usage Policy violations, account creation from an unsupported location, and Terms of Service violations. It publishes an appeal route for users who believe an account was wrongly suspended or terminated, reached by logging in with the affected account, along with self-serve data export and account deletion for banned users, and a review request for organizations placed on hold. Anthropic notes that the data a banned user can access and export may be restricted depending on the violation associated with the account. That describes the published process, and nothing about whether every measure is appealable or how an appeal resolves. No OpenClaw-specific enforcement outcome is established anywhere in the audited record.
Sources: Anthropic, Claude Code Docs, Legal and compliance, accessed 3 August 2026 · Anthropic Help Center, Safeguards warnings and appeals, accessed 3 August 2026 · Anthropic, Consumer Terms of Service, effective 8 October 2025
Verdict
Pro and Max increase native subscription capacity, priority access and weekly-limit headroom, and both include Claude Code and Cowork. API access stays a separate product with separate payment and a separate contract. Across the Anthropic pages audited on August 3, 2026, none publishes a Max-specific expansion of third-party permission, and none resolves OpenClaw’s status in either direction.
Establish the route and the active credential before choosing a tier. Working code establishes capability, billing documentation identifies a meter, and only Anthropic’s policy states Anthropic’s position.
FAQ
Was Claude Pro or Max universally cut off from OpenClaw?
Not as a published universal rule. Anthropic’s current Agent SDK page says third-party app usage still draws subscription limits for now. Its login guidance allows certain third-party tools at Anthropic’s discretion. Neither names OpenClaw, and no audited Anthropic page publishes an OpenClaw-specific decision in either direction.
Does Max give broader third-party access than Pro?
No reviewed Anthropic page says so. Max documents five or twenty times Pro’s usage per session plus priority access. Among the factors those pages name for third-party tool use are subscriber status, usage-credit enablement, honest client identity and Anthropic’s discretion. The plan tier is not among them.
Does a Pro or Max subscription include the Claude API?
No. Anthropic states that a paid Claude subscription does not include access to the Claude API or Console, and the Pro page repeats that API usage through Console is excluded. Console access requires separate setup and payment. Paid-plan Usage Credits are billed at standard API rates but are not API entitlement.
Does claude -p use my Pro or Max subscription?
Anthropic’s current Agent SDK page says claude -p usage still draws subscription limits for now, after a proposed change was paused in June 2026. That statement identifies a meter in the Agent SDK context. An effective API key changes the billing account, and permission questions for third-party controllers remain separate.
What happens if an API key is set while I am logged into a subscription?
Anthropic documents that Claude Code prioritizes an environment-variable API key over an authenticated subscription and bills that key’s API account. It also documents a setup-time confirmation and a post-login notification when both are active. Run /status to see which method is currently active.
Will Anthropic warn me before paid-plan Usage Credits start charging?
For the ordinary post-limit flow, yes. Anthropic documents prior enablement, prepayment, a notification on approaching and reaching the included limit, and a confirmation before continuing. Its discretionary metering of certain third-party tools is separate, and the trigger and prompt behavior for that classification are not publicly mapped.
What does claude setup-token change when used with OpenClaw?
Anthropic mints a one-year OAuth token, prints it and saves no copy. Pasting it into OpenClaw places that bearer credential in OpenClaw’s own store. From there the project implements two branches: its own runtime, or a configured CLI backend. Permission, meter and governing contract remain unresolved publicly.
Can a suspended or terminated user appeal?
Anthropic publishes an appeal route for users who believe an account was wrongly suspended or terminated, reached by logging in with the affected account. Banned users can also export data or delete the account, though exportable data may be restricted. This describes the published process, not any outcome.
Methodology
This is a Tier C briefing, built from documents and fixed source code rather than testing. Anthropic-controlled pages and fixed OpenClaw source files were opened directly on August 3, 2026, and are listed in full at the end of this page. Where a page displays a relative update date, that is recorded as shown rather than converted into a fixed date.
FSR opened no account for this briefing, ran none of the routes, observed no billing screen and reproduced no error. Source-code inspection proves what the audited tree contained at one snapshot, not what any installation is configured to do, and not permission. OpenClaw’s live documentation is mutable and is cited as a page state on the access date, while historical claims are cited to fixed commits so they remain checkable after the page changes.
Absence claims are bounded. Where this briefing says something is not published, it means it was not found in the Anthropic-controlled source set audited on August 3, 2026. That search did not cover company posts on external platforms, private correspondence or unpublished agreements. Several external AI systems widened the search for candidate sources; none of their output was treated as evidence, and agreement among them was not treated as corroboration. Anthropic had not responded to FSR’s route-specific Support request as of August 3, 2026.
Source: FSR editorial method, applied 3 August 2026.
Correction and update log
August 3, 2026 — Material correction and full replacement. Removed the original universal-cutoff framing; separated API-key, native OAuth, local claude -p, setup-token, Agent SDK, paid-plan Usage Credits and OpenClaw-specific routes; removed unverified metrics, causal claims, motive claims, competitor comparisons and predictions; added the tested and untested boundary and disclosed the pending Anthropic Support request.
Contact us
Correction, additional primary source, or a route we should audit next? Anthropic’s response to our route-specific request will be appended here when it arrives.
[email protected]Related FSR briefings
Tier B briefings are hands-on tested. Tier C briefings are document-first.
-
TIER B
Claude Code Review: 5 Costly Mistakes Every Developer Makes
The Claude Code route in practice, from a review focused on costly setup mistakes.
-
TIER B
Claude Fable 5 Is Back, But Its Usage Meters Do Not Agree
What happens when the surfaces reporting your usage do not agree with each other.
-
TIER B
Odysseus Review: Where PewDiePie’s Self-Hosted AI Workspace Actually Sends Data
The custody question in another agent workspace: where the data actually goes.
-
TIER C
Claude Pro, Max, Team, and Enterprise: The Control Boundaries Behind the Plan
The organizational counterpart to this briefing: plan, surface and execution controls for Team and Enterprise buyers.
-
TIER C
Claude Fable 5 Pricing: Anthropic Tells You the Token Price, Not What Your Plan Includes
The same gap on the model axis: a published token price does not tell you what your plan includes.
-
TIER C
Kimi K3 for Technical Buyers: Which Route You Take Changes What You Bought
Route before tier, applied to a different vendor, with the same purchase consequences.
Future Stack Reviews publishes independent structural audits of AI and SaaS products. This briefing is document-first and reports no hands-on testing. It is not legal, financial or compliance advice, and it does not state whether any specific use is permitted. Product terms, prices and policies change; verify against the vendor’s current pages before acting.
Sources opened 3 August 2026. Last updated 3 August 2026.
Source list
Anthropic-controlled, all opened 3 August 2026
A01 Log in to your Claude account, 19 May 2026
A02 Use the Claude Agent SDK with your Claude plan, 16 June 2026
A03 Legal and compliance
A04 Authentication
A05 Agent SDK overview
A06 What is the Max plan?
A07 Commercial Terms of Service, effective 17 June 2025
A08 Consumer Terms of Service, effective 8 October 2025
A09 Manage usage credits for paid Claude plans, 18 May 2026
A10 Safeguards warnings and appeals
A11 What is the Pro plan?, 10 June 2026
A12 Paid subscription versus API and Console, 16 March 2026
A13 Use Claude Code with your Pro or Max plan, 11 June 2026
A14 Manage API key environment variables in Claude Code, 5 May 2026
A15 Buy usage bundles, 18 May 2026
OpenClaw-controlled, all opened 3 August 2026
B01 commit 37301cb, 4 April 2026
B02 commit d378a50, 6 April 2026
B03 OAuth documentation
B04 snapshot e812630f
B05–B11 the seven source files cited in section 04, all pinned to e812630f
Stay with the review desk
Choose a channel to keep reading.
Share this review