Model Response Version Does Not Match the Specified Version?

😱 Why Does the Model’s Answered Version Not Match the Version You Selected?

Why does this topic deserve a standalone document? Because I really do not want to keep answering this anymore!!!

Let’s start with the conclusion: No model knows its exact version. Any model that can state its exact version is being manually faked!!! (Except in online-connected scenarios)

Let’s look at a test:

---

Easily Exposing Model Version Lies!

First, in Chatbox, we selected claude-opus-4-5-20251101 and claude-opus-4-5-20251101-thinking one after the other. The LLM accurately stated its own version — so “smart,” so “powerful”!

Let’s take a look at the prompt Chatbox sends to the API:

So that’s what’s going on: as long as the content of the system role includes the current model version, the LLM can answer with pinpoint accuracy, creating the illusion that it is both very smart and “genuine.”

Next, let’s look at Cherry Studio (the reliable, honest model-student version)

It’s clearly claude-opus-4-5-20251101, so why does it say it is Claude 4? Explosive. Faith shattered. Did it get dumber? Or is the seller being dishonest?

Let’s inspect this request. The content for the system role is 1111 (ignore this boring system prompt — I wrote it casually last time. Cherry Studio’s default system prompt is empty and is not sent).

---

Only two tools are listed here, but they actually represent two different approaches.

1. Cherry Studio is more thoroughly open source and pursues a more native model experience, so it does not add preset interfering prompts. Other examples in this category include various coding CLIs, such as Claude Code CLI. Their system prompts are mainly coding constraint rules, and of course they would not go out of their way to fake the model name for something this trivial.
2. Chatbox is open source, but it launched its own expensive membership plan and even claims to have its own models. If it did not cooperate with this preset prompt, wouldn’t its cover be blown? Similar examples include some tavern-style chat apps with membership-based pricing that also call a deer a horse — saying A is B. They’re trying to cut costs and make huge profits, so savvy users should judge for themselves.

---

4All API - One-stop AI model API aggregation platform
Website: https://4allapi.com
API Base: https://api.4allapi.com