OpenClaw 12 min read

Is OpenClaw Dead? 2026 Status, Safety, Cost, Claude Terms and Alternatives

Is OpenClaw dead? No: 2026.11.4 shipped 16 Nov 2026, its 22nd release in 30 days. Is it free, safe, open source or worth it? Straight answers, checked.

Shabnam Katoch

Shabnam Katoch

Growth Head

Is OpenClaw dead? Status as of 16 November 2026: latest release 2026.11.4, 22 releases and 16,569 commits on main in 30 days

Status as of 16 November 2026

What we checkedResultHow we counted
Latest release2026.11.4, 16 Nov 2026, marked LatestGitHub releases page
Extended-stable line2026.9.37, 10 Nov 2026GitHub releases, npm extended-stable tag
Releases, last 30 days22: 10 stable, 4 extended-stable, 8 betaEvery GitHub release dated 17 Oct to 16 Nov; npm lists the same 22
Commits to main, last 30 days16,569 from 370 author namesgit log on a clone of main, by commit date
Issues opened, last 30 days4,805, of which 342 are titled "Update failure"GitHub issue search by created date
Open issues and pull requests4,910 issues, 3,359 pull requestsGitHub Issues and Pull requests tabs
Stars and forks395k stars, 82.9k forksGitHub repo page
Published security advisories722GitHub Security tab
Licence and stewardMIT, OpenClaw Foundation, a 501(c)(3)LICENSE and README
Claude subscriptionsNo ban by name; Anthropic limits plan logins in third-party appsAnthropic legal page and help centre, OpenClaw docs

Checked 16 November 2026. The window runs from 17 October 00:00 UTC to 16 November 2026, and the commit count stops at the newest commit on main, 10:18 UTC on 16 November.

Picture this: your OpenClaw agent skipped the morning briefing again, and the first headline you find says the project is dead.

Do I fix this, or start over with something else?

That depends less on OpenClaw than on what you want to own. Below are the "Is OpenClaw" questions Google suggests, each answered in two or three sentences first, with the evidence underneath.

Is OpenClaw dead? No, and it isn't close

No. A dead project does not publish 22 releases in 30 days. Between 17 October and 16 November 2026, OpenClaw shipped 10 stable releases, 4 extended-stable builds and 8 betas, and 370 different author names landed commits on main.

The stable line moved from 2026.10.5 on 18 October to 2026.11.4 on 16 November, and our OpenClaw latest version guide tracks what changed in each build. Its creator, Peter Steinberger, is still the most frequent author on main: 7,468 of the window's 16,569 commits carry his name.

Is OpenClaw dead? 22 OpenClaw releases between 17 October and 16 November 2026 across the stable, extended-stable and beta channels

ChannelReleases, 17 Oct to 16 NovCurrent versionWhat it is for, per OpenClaw's docs
Stable (npm latest)102026.11.4 (16 Nov)Recommended for most users
Extended-stable42026.9.37 (10 Nov)A trailing, package-only line that never updates itself
Beta82026.11.5-beta.4 (16 Nov)Where builds are usually vetted before stable

Checked 16 November 2026.

If you run the quiet channel, read the 2026.9.37 notes closely. They say it was "published from a retired maintenance line by explicit maintainer decision", so check what the extended-stable tag points to before you pin.

Because the best-known "dead" verdicts were never about the code. They were about cost, trust and noise, and all three are real.

Forrester's "OpenClaw Is Dead, Long Live OpenClaw" by Charles Betz (9 April 2026) is about money. He describes an Anthropic email saying tools like OpenClaw would no longer be covered by his Claude Max plan. His daily assistant became a metered cost of $25 to $50 a day by his count, so he routed it to a cheaper model through OpenRouter and kept going.

A Medium post titled "OpenClaw is Dead" by Mehul Gupta (14 May 2026) makes the trust case. Even it concedes the project is still active, arguing that a project "can remain alive in code while slowly dying in reputation, trust, and adoption."

Then there is the noise. Of the 4,805 issues opened on GitHub in our window, 342 carry titles like "Update failure: database-schema-preflight (2026.11.3)", naming the update step that failed and the version. To us that reads as a busy release train, not a collapse, and our decoder entry for update failure: reconcile abandoned explains one of those verdicts.

Old Macs add to the feeling. The native OpenClaw.app now requires macOS 26.2 (Tahoe) or later, while the CLI and Gateway are supported back to macOS 13.5, so a Mac on an older system can run OpenClaw but not open its app.

"Is OpenClaw dead?" is really a question about trust. The release data settles the code question. Trust is the part you have to judge for yourself.

Is OpenClaw free?

Yes, the software is free. OpenClaw is MIT-licensed, and its README says the project "has no paid tier, hosted service, or token." Running it is not free, because you pay for the model, for a machine that stays on, and in time.

What you pay forWhat it meansWho gets paid
The softwareMIT licence, no paid tierNo one: $0
The modelAPI usage by the token, or the limits of a plan you already pay forYour model provider
The machineYour own computer, or a server that stays on around the clockYou, or a host
The upkeep10 stable releases in our 30-day window to read, apply or skipYour time

Checked 16 November 2026.

That last row is the one people forget. Ten stable releases in 30 days is roughly one every three days (our calculation), and each is a small decision: update now, wait, or pin a quieter channel. For real monthly numbers by model, see our breakdown of OpenClaw API costs.

Can you still use a Claude subscription with OpenClaw?

There is no Anthropic ban on OpenClaw by name in any Anthropic terms or docs we could find. What Anthropic's rules do say is that third-party developers may not offer Claude login inside their apps or route requests through Free, Pro, Max or Ultra plan credentials, and OpenClaw's own docs recommend an Anthropic API key for shared or production agents.

Here is what each side says, in its own words.

SourceWhat it says
Anthropic, Claude Code legal and compliance pagePlan sign-in (OAuth) is "intended exclusively for purchasers of Claude Free, Pro, Max, Ultra, Team, and Enterprise subscription plans" and supports "ordinary use of Claude Code and other native Anthropic applications". Developers "should use API key authentication", and Anthropic "does not permit third-party developers to offer Claude.ai login into their own applications, or to route requests through Free, Pro, Max, or Ultra plan credentials on behalf of their users". It may enforce this "without prior notice".
Same pageThe rules do not stop "an end user from signing in to the unmodified Claude Code binary with their own Claude subscription", and advertised limits "assume ordinary, individual usage of Claude Code and the Agent SDK".
Anthropic Consumer Terms (effective 8 October 2025)Automated access "through a bot, script, or otherwise" is barred, with one carve-out: "Except when you are accessing our Services via an Anthropic API Key or where we otherwise explicitly permit it".
Anthropic help centre, 15 June 2026 update"Claude Agent SDK, claude -p, and third-party app usage still draw from your subscription limits."
OpenClaw docs, Anthropic provider pageTwo routes: an API key, or "reuse an existing Claude Code login through the installed executable on the same host". API-key billing "is preferable for shared automation or predictable production spend."

Checked 16 November 2026.

So where did "Anthropic banned OpenClaw" come from? In April 2026, press reports quoted an Anthropic email telling subscribers that their plan limits would no longer cover third-party tools such as OpenClaw, with that use billed as extra usage instead. We could not find that email on Anthropic's own pages, and the help centre's 15 June update now says third-party app usage still draws from subscription limits.

OpenClaw's docs also describe a setup-token option: you run claude setup-token, paste the long-lived token into OpenClaw, and OpenClaw calls Anthropic's API with it directly. Set that next to Anthropic's line that developers "may not collect, store, or intermediate Claude.ai credentials or session tokens", and you can see why we would not rest an always-on agent on it.

Claude in OpenClaw: an API key versus reusing a Claude Code login, and what Anthropic's rules say about plan logins in third-party apps

For an always-on agent, use an API key. It is what Anthropic tells developers to use and what OpenClaw recommends for production, and a runaway loop then meets a spend limit you set instead of your personal plan.

If you already pay for Max or Ultra, part of that bill may be covered. Since 7 October 2026 those plans include monthly API credits for the Claude API ($100 on Max, $200 on Ultra), claimed into a linked Claude Console organization, and any API key in that organization draws on them until they expire at the end of each billing cycle. Our guide to bring your own key AI explains how keys and billing fit together.

Is OpenClaw safe?

Safe enough for one person or a trusted team who hardens it, and not safe as one shared agent for people who do not trust each other. That is OpenClaw's own trust model, and its defaults assume a trusted operator: tools run on the host for the main session unless you set up sandboxing.

The record cuts both ways. GitHub lists 722 published security advisories for OpenClaw, a sign of a large attack surface and of maintainers who publish what they fix. Earlier this year, CVE-2026-25253 (rated 8.8, High, in its CVE record) let a crafted link make versions before 2026.1.29 connect to a gateway address taken from the link, without asking, and send a token.

Is OpenClaw safe? One Gateway is one trust boundary: supported for one operator or a trusting team, not for mutually untrusted users

DefaultWhat it meansWhat we would do
One trust boundary per GatewaySupported for one operator or a team that trusts each otherRun separate Gateways for separate teams
Tools run on the hostThe main session acts directly on your machineTurn on sandboxing
Host exec without prompts for a trusted single operatorCommands run with no approval stepTighten exec approvals before adding channels
Unknown DM senders must pairNew contacts need your approval firstKeep it on
Session tools reach every agentSession visibility defaults to allNarrow it if agents differ in trust

Checked 16 November 2026.

Before you connect other people or open the Gateway beyond your own machine, run the built-in audit and apply its safe fixes:

openclaw security audit --deep
openclaw security audit --fix

Then vet every skill before you install it, and keep the Gateway private until you have worked through OpenClaw's exposure runbook. For past incidents and the mitigations in one place, see our OpenClaw security breakdown.

Is OpenClaw still a thing in 2026, and is it still relevant?

Yes, by a wide margin. OpenClaw has 395k GitHub stars and 82.9k forks, well ahead of Hermes Agent's 264k stars, and npm counted 12,216,646 downloads in the 30 days to 15 November, about twice May's 5,812,288 (our calculation).

npm counts every install and update, including automated ones, so treat that as a trend line, not a head count.

Who pays for it says something too. The OpenClaw Foundation lists donors including Amazon, OpenAI, Red Hat and the University of Michigan, with infrastructure support from GitHub, NVIDIA and Vercel, and NVIDIA's own NemoClaw installs OpenClaw by default.

Is OpenClaw good, and is it worth it?

OpenClaw is good at what it was built for: a personal agent on your own hardware that lives in the chat apps you already use, backed by the biggest community of any open-source agent we track. It is worth it if you enjoy running infrastructure and will keep up with a release train that shipped 22 builds in 30 days, and not worth it if you only want what the agent does.

Here's the thing. The software costs nothing, so the real price is attention: reading release notes, choosing when to update, hardening the Gateway, and owning the fix when something breaks at night.

Be honest about which of those you will actually do. If the answer is "all of them, happily", OpenClaw is still the default choice, and nothing in its release data says otherwise.

If the answer is "the agent, yes; the server, no", that is the gap BetterClaw was built for. BetterClaw does not host OpenClaw: it runs its own agents, each in a sandboxed container, on your own model key with zero markup. Plans start at $19 a month. See BetterClaw plans

Is OpenClaw open source?

Yes. The code is on GitHub under the MIT licence, copyright the OpenClaw Foundation, an independent 501(c)(3) that employs the core team and signs releases. The README is plain about ownership: "OpenAI is a donor, not an owner."

Open source also means you can check what it sends home. By default that is only a daily version check, anonymous feature statistics are opt-in, and update.checkOnStart: false switches both off.

One caveat: skills and plugins you add from ClawHub are separate code from separate authors. "OpenClaw is open source" does not make every add-on safe.

Is OpenClaw agentic AI?

Yes. OpenClaw is an agent runtime: give it a goal and it decides the steps and acts, running shell commands, reading and writing files, using a browser and sending messages in your chat apps. Its built-in scheduler can also wake an agent at a set time, so work starts without you typing first.

The model supplies the reasoning, and OpenClaw supplies the hands, the memory and the clock. That split is why safety matters here, because the actions are real even when the reasoning is wrong. If the term is new to you, start with what agentic AI means.

Is OpenClaw a harness?

If you mean the software wrapped around a model, yes. OpenClaw is not a model: it is the gateway and runtime that give a model tools, memory, chat channels and a schedule, and you choose the model provider.

Its docs draw the lines more finely. They separate four layers, and the agent runtime, the loop that actually executes a turn, is swappable: OpenClaw's built-in runtime, Codex, GitHub Copilot or a local Claude CLI.

LayerExamples from OpenClaw's docsWhat it decides
Provideranthropic, openai, github-copilotHow OpenClaw signs in and names models
Modelclaude-opus-5, claude-haiku-5-5Which model answers the turn
Agent runtimeopenclaw, codex, copilot, claude-cliThe loop that executes the turn
ChannelDiscord, Slack, Telegram, WhatsAppWhere messages come in and go out

Checked 16 November 2026.

So the short answer: OpenClaw is the framework, and even the loop inside it can be swapped.

Is OpenClaw dead for you? When to stay, when to move

The project is alive, so the useful question is whether it is alive in your setup. That mostly comes down to one thing: do you want to run the server?

Your situationWhat we would doRead next
It works and you like running itStay: run the audit, use an API key for always-on agents, choose a release channel on purposeThe safety checklist above
You like OpenClaw, not the server choresStay on OpenClaw, on a managed hostOpenClaw hosting compared
You need stronger isolation around OpenClawLook at NVIDIA's NemoClaw, which runs agents inside OpenShell sandboxes, OpenClaw by defaultBetterClaw vs NemoClaw
You want a different self-hosted agentTry Hermes Agent, which can import your OpenClaw settings, memories and skillsBetterClaw vs Hermes Agent
You want the agent, not the serverUse a managed agent platformBetterClaw vs OpenClaw

Checked 16 November 2026.

Is OpenClaw dead for you? Keep self-hosting and harden it, add NemoClaw sandboxing or switch to Hermes Agent, move to a managed OpenClaw host, or use a managed agent platform

OpenClaw is not dying. It is doing what fast infrastructure projects do, shipping faster than most people want to operate it.

So the question was never whether the lobster is alive. It is whether you want to be the one who feeds it at 2am.

Not sure whether to stay or move?

If you are weighing OpenClaw against a move, an outside view helps. Our AI automation audit is a 30-minute call, followed by a written report within 5 business days. Book an AI automation audit

Frequently Asked Questions

Did Anthropic ban OpenClaw?

Not by name in any Anthropic terms or docs we found. Anthropic's rules bar third-party apps from offering Claude login or routing requests through Free, Pro, Max or Ultra credentials, and tell developers to use API keys. Its help centre says, in a 15 June 2026 update, that third-party app usage still draws from your subscription limits.

What is the latest version of OpenClaw?

OpenClaw 2026.11.4, published on 16 November 2026 and marked Latest on GitHub. The extended-stable line is at 2026.9.37 (10 November), and the newest beta is 2026.11.5-beta.4. Check yours with openclaw --version.

Is OpenClaw still maintained?

Yes. In the 30 days to 16 November 2026 it published 22 releases and took 16,569 commits on main from 370 author names, and the OpenClaw Foundation employs the core team. Expect a busy tracker: 4,805 issues were opened in the same window.

Who owns OpenClaw now?

The OpenClaw Foundation, an independent 501(c)(3), develops it in the open, employs the core team and signs releases. Donors include Amazon, OpenAI, Red Hat and the University of Michigan, and the README says plainly that OpenAI is a donor, not an owner. Peter Steinberger, who built it, was still the most frequent author on main in our 30-day count.

What is the best alternative to OpenClaw?

It depends on what you are leaving. Hermes Agent suits people who want to keep self-hosting and can import your OpenClaw settings, memories and skills, a managed host keeps OpenClaw without the server work, and a managed platform gives you the agent with no server at all. Our comparison of nine OpenClaw alternatives covers each.

Should a business run OpenClaw?

Only as one Gateway per trusted team. OpenClaw's trust model supports one operator or a mutually trusting team per Gateway and does not support one shared agent for users who do not trust each other, so separate teams need separate Gateways. Budget for someone who owns hardening, updates and incident response.

Want to skip the setup?

BetterClaw does this in 60 seconds. No Docker, no config files.

Start free
Tags:is openclaw deadopenclaw is deadis openclaw freeis openclaw safeis openclaw open sourceanthropic ban openclawopenclaw latest versionopenclaw alternative
Shabnam Katoch

Written by

Shabnam Katoch

Growth Head

Shabnam leads growth at BetterClaw and has spent the past three years building organic distribution for AI-native products. She writes about no-code agent builders, content strategy, and what it actually takes to grow without an ads budget.

Share this article
Was this helpful?