On April 4, 2026, Anthropic cut off Claude Pro and Max subscribers from using OpenClaw and other third-party agent harnesses on their plans. The flat-rate “run autonomous agents all month” deal that powered a generation of builders collapsed overnight into pay-as-you-go billing — with power users reporting bills 10–50x higher than their old subscription.
The message was clear: the big providers’ “unlimited” plans were never built for agents. Ours are.
Today we’re making every paid Shogo plan unlimited — and unlike Claude’s subscriptions, autonomous agent workloads aren’t a banned edge case here. They’re the entire point.
What Anthropic actually changed
If you ran OpenClaw on a Claude Pro ($20) or Max ($100–$200) subscription, you were getting open-ended agent usage for a flat monthly fee. As of April 4, that’s gone:
- Claude’s consumer subscriptions no longer cover third-party agent harnesses like OpenClaw.
- To keep going, you move to pay-as-you-go “Extra Usage” or a metered API key — billed per token, not per month.
- A later partial walk-back only restores agent use through a small fixed monthly credit, after which you’re back to API rates.
It’s a rational move for them — autonomous agents burn tokens fast, and a flat fee doesn’t cover it when the underlying model is expensive. But it leaves builders who depend on agents stuck with metered bills that balloon the moment real work starts.
Unlimited that’s actually built for agents
Every paid Shogo plan is now unlimited usage — no draining a fixed monthly credit pool, no per-task meter ticking against you. Higher tiers simply get more headroom:
- Basic — Unlimited usage.
- Pro — Unlimited usage with higher limits, per seat.
- Business — Our highest usage limits, per seat.
- Enterprise — Uncapped.
Autonomous agents driving tools, editing files, and running long jobs are exactly what these plans are sized for — not something we’ll email you about banning on a Friday night.
And Shogo beats OpenClaw
This isn’t just a better deal — it’s a better agent. Across three independent, harness-neutral evaluations, Shogo powered by Hoshi 1.0 beats OpenClaw running Claude Haiku 4.5 on every one:
| Eval | Shogo + Hoshi 1.0 | OpenClaw + Haiku 4.5 |
|---|---|---|
| ClawBench Core v1 (agentic, 0–1) | 0.844 | 0.797 |
| ClawEval (1,220 checkpoints) | 85.2% | 84.7% |
| Aider Polyglot (83 tasks) | 83 / 83 (100%) | 81 / 83 (97.6%) |
Tool use, deterministic output, and multi-language coding — three different muscles, three wins. Full methodology is in the Hoshi 1.0 launch write-up.
12x* the work per dollar
The reason we can offer unlimited agent usage where Claude’s plans pulled back: Shogo runs Hoshi 1.0, our own model. On the identical 83-task coding set, the same money goes about 12x further than Haiku-class pricing — at matching task completion:
| Shogo + Hoshi 1.0 | OpenClaw + Haiku 4.5 | |
|---|---|---|
| Cost per task | $0.005 | ~$0.06 |
| Total (83 tasks) | $0.41 | ~$5.2 |
When each unit of work costs roughly a twelfth as much, “unlimited” stops being a loss-leader and starts being the obvious default. The full derivation is in our work-per-dollar benchmark.
How the windows work
We’re not going to pretend there’s no ceiling — that’s the kind of “unlimited” that ends in a Friday-night email. Here’s the honest version:
- Usage is unlimited within two rolling windows — a 5-hour burst window and a 7-day weekly window — that reset on their own schedule.
- Higher tiers carry higher windows; Enterprise is fully uncapped.
- If you do max out a window, you either wait for it to reset or turn on optional overage to keep going at metered rates — your choice, with a cap you control.
No OAuth ban. No surprise migration to pay-as-you-go. Just generous limits sized for the way agents actually work.
From the team
“Anthropic banning agents from their unlimited plans told you everything about who those plans were for,” said Guru Angisetty, CEO and Co-Founder of Shogo. “Ours are built for the opposite — the person who wants an agent working all day. Because we run Hoshi 1.0, we can make that unlimited and still come out ahead.”
“OpenClaw on Haiku is the stack we already beat on every eval,” added Russell LaCour, CTO and Co-Founder. “Now we beat it on price too: unlimited agent usage, no metered surprises, and roughly 12x the work for the same dollar.”
About Shogo
Shogo is the platform where AI agents build the software you need instead of replying with text. Describe what you want, and a Shogo agent — powered by Hoshi 1.0 — builds it: a live dashboard, an automated workflow, a visual interface wired to your real tools. Start unlimited on a paid plan, or deploy your first agent in minutes on the free tier.
*12x is based on measured cost-per-task on the 83-task Aider Polyglot set — Hoshi 1.0 at $0.005/task vs Claude Haiku 4.5 at ~$0.06/task, at matching task completion. Hoshi’s cost is metered exactly from per-task token logs; Haiku’s is estimated by applying its list rates to the same measured token workload. See the work-per-dollar benchmark for the full method.
Unlimited usage is subject to fair-use rolling 5-hour and weekly windows; higher tiers carry higher windows and Enterprise is uncapped. If you max out a window, you can wait for it to reset or enable optional overage at metered rates, with a cap you control.