Misleading pricing, model substitution (spoofing), and hidden x8.3 markups
In the product description, the seller promises an honest rate of 1M = $1 (list price), Fable 5.1 and Opus 5.5 models for working in Claude Code, and near-free prompt caching. There is not a single word about any "channel" system in the listing description.
The truth only comes out in the user dashboard after you pay. It turns out that the base "Standard" plan at the advertised list price is a crippled economy tier "for chatbots and translations." In the description of the neighboring "Pro" channel with a x2.8 markup, the seller openly writes: "Opus 5.5 and Sonnet 5.5 without substitution." This means that on the base channel—at the price promised in the description—models are officially substituted! And to get the advertised Fable 5.1 model, a working system prompt, full prompt caching, and honest token accounting as at Anthropic, you have to switch to the "Max" channel with a x8.3 markup!
On the base channel, the response quality is terrible: "Opus 5.5" at the maximum xhigh level, which should take a very long time to think, spits out a weak answer in about 30 seconds and looks nothing like the original model. Moreover, their routing gives itself away by dumping full raw chains of thought, even though the real Opus 5.5 has encrypted thoughts that are hidden by default (with at most a brief summary available). Returning full raw thoughts directly indicates that traffic is being routed either through the old Haiku 4.5 or through cheap Chinese models.
Even worse, the seller's broken gateway forcibly injects garbage messages into the conversation history—"OK" from the assistant and a dot "." from the user. The model sees this garbage in its context, gets confused, and wastes your paid tokens reasoning along the lines of: "The single "." might be a way to prompt me to continue (since I just said "OK" previously without answering)".
Chat support initially tried to gaslight me, claiming that raw thoughts on Opus 5.5 are normal and that I had typed the "." and "OK" into the chat myself. When I cornered them with facts, support backtracked and officially admitted both points: that Opus 5.5 does not return thinking text by default, and that "." and "OK" are injected into the context by their own channel.
