Claude by Anthropic
A family of foundational AI models
Start new thread
fmerian

Running OpenClaw with Claude subs is dead. Now what?

by•

As of April 4th, Claude subscriptions no longer cover usage on third-party tools like @OpenClaw.

If you were running your agent with @Claude by Anthropic, curious what option you'd pick: would you pay the extra bundle? pick another provider? or use an alternative like @KiloClaw or @Zo Computer?

1K views

Add a comment

Replies

Best
You can always self-host a model with e.g. Ollama, or use a dedicated inference provider like Nebius (either just for the API calls, or they provide sandboxed OpenClaw instances). If you're going to use a claw-like agent loop you definitely shouldn't be using OpenClaw directly just because of the terrible security and guardrails. Find something open source you can verify, run it inside a container, and don't ever give it permissions to impact anything important. Qwen3.5-coder is supposed to be quite good at the right kinds of tasks, although needs the fix from Unsloth (see https://www.reddit.com/r/LocalLL... ), and I've heard good things about the Kimi models as well.
fmerian

Qwen3.5-coder is supposed to be quite good at the right kinds of tasks (...) and I've heard good things about the Kimi models as well.

yes! and according to @PinchBench.com, @MiniMax and @Arcee's latest models are worth a spin, too

J.D. Salbego

@fmerian I'm a die-hard opus 4.6 extended thinker for everything. Why use a college grad's brain when you have Einstein's brain at your fingertips? But yes, it's incredibly expensive, but when you calculate the output value vs. the costs, it's a no-brainer.

fmerian

@jdsalbego are you using Opus 4.6 for both your agentic and coding tasks? heard great things about @MiniMax M2.7 and @Arcee's Trinity-Large-Thinking

J.D. Salbego

@fmerian Yeah, across every work stream and department. Im very loyal lol

ABS Astreon
for those of you using Claude code, don't forget to collect your free credits under the usage page, make sure the toggle is checked before you click the claim button. I'm sure most of you know this, just a reminder.
Pashupathi Mali

This feels like an inevitable shift — subscriptions were never meant for production-grade or always-on agents.

Curious how many people were depending on a single model/provider though — changes like this can break systems overnight.

Feels like moving toward multi-model setups is becoming necessary now.

fmerian

@pashupathi exactly

Matt

I don't know if they can even do something like that tbh, perhaps it's the way we are harnessing Claude Code x OpenClaw (we don't mess with authentication, rather function more as a developer tool, like a terminal). No changes on our side at @Toone

Steriani Karamanlis

the switching decision is harder than it looks. same model, different vendor, prices vary up to 6x once you normalize for context window, caching availability, and output weight. most people pick a provider once and never revisit it. but the market has shifted enough in the last few months that what made sense 6 months ago probably isn't optimal today. worth running the actual numbers before committing to a new stack.

Podrick Clawdius

I am using MiniMax M2.7 and Codex 5.3 for general Agerntic tasks and Coding respectively - MiniMax Coding Plan for $20 a month and codex via my $20 monthly plan, never had an issue with Rate limits.

fmerian

thanks for the feedback; heard great things about @MiniMax.

this thread compares M2.7 with Opus 4.6 and tl,dr: it delivered 90% of the quality for 7% of the cost. definitely worth a spin for agentic tasks imho.

Sayanta Ghosh

Most of the stuff that I tried on OpenClaw can easily be done on Cowork. I shifted to Cowork a long time back.

What use cases am I missing that MCPs and Cowork is unable to do?

Priyanshu Singh

Feels like this will push people to rethink their stack more than just swap providers.

A lot of setups were built assuming bundled access. Now it’s more about control vs convenience.
I’ve seen some people move toward simpler workflows instead of chaining too many tools.

Luca Ardito

I’m seeing more builders realize that “cheap enough to test” and “safe enough to run daily” are two different decisions.

Would love to hear if anyone here found a setup that feels reliable for everyday work without turning into a full infra project.

12
Next
Last