Max 20 Plan Delivers Claude 3 Opus Instead of Claude 4 Opus
TL;DR
I’m paying \$200 per user per month for the “Claude Code Max 20” plan, which explicitly advertises and promises Claude Opus 4.
Yet the model actually running—and acknowledged by Claude itself—is Claude 3 Opus (model ID claude-3-opus-20240229).
---
📦 Plan Details
| Item | Value |
| -------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Plan | Claude Code Max 20x |
| Price | \$200 / user / month |
| Advertised model | Claude Opus 4 |
| Marketing claims | “Deep coding at terminal velocity with Claude Opus 4”<br>“Optimized for code understanding with Claude Opus 4”<br>Max plan gives “the most access to Claude Opus 4” |
<details>
<summary>See the Screenshots of marketing and plan page</summary>
- Max 20 plan card (price & “access to Claude Opus 4”)
<img width="496" height="367" alt="Image" src="https://github.com/user-attachments/assets/49edc5b1-3bee-4bd4-b381-01ce5f951099" />
- Feature list claiming Claude Opus 4 powers code understanding
<img width="568" height="215" alt="Image" src="https://github.com/user-attachments/assets/20913d0d-fe96-4a5c-b4ba-36ae351e4678" />
- Header: “Claude Code embeds Claude Opus 4”
<img width="936" height="249" alt="Image" src="https://github.com/user-attachments/assets/61611a8b-cf5a-4e2f-8330-155c31ce1a5b" />
</details>
---
🔎 Observed Behavior
- Generated code banner
``
This codebase was generated by Claude Code (Anthropic Claude 3 Opus, model ID: claude-3-opus-20240229)
- Direct model query
``
> Which model am I currently using?
⏺ You are currently using Opus (specifically Claude 3 Opus, model ID: claude-3-opus-20240229).
Claude further clarifies: “Claude 4 hasn’t been released yet—you are on Claude 3 Opus.”
<img width="723" height="479" alt="Image" src="https://github.com/user-attachments/assets/35f7fb4e-2baa-440d-942f-2de88b16dd00" />
<img width="270" height="44" alt="Image" src="https://github.com/user-attachments/assets/3aadeeb4-5209-45b8-a9d5-681d51da9e08" />
HOWEVER, when I check which model I am using, it incorrectly provides me with Claude 4 Opus
<img width="941" height="187" alt="Image" src="https://github.com/user-attachments/assets/a1de0c26-c11f-414c-9b0e-d9f3e9e2ec38" />
---
🛑 Why This Is a Critical Issue
- Contractual mismatch — I’m paying for access to a higher-tier model (Claude 4).
- Misleading product claims — All marketing, plan copy, and in-app messaging insist Claude 4 is included and live.
- Potential backend misconfiguration — Some (or all) Max 20 tenants may be incorrectly routed to Claude 3.
- Competitive risk & trust — Customers rely on accurate model tiering for capability and security evaluations.
---
✅ Expected Behavior
- Requests made under a Max 20 subscription should be served by Claude Opus 4 immediately.
- The model banner and
model_idshould reflect Claude 4 (e.g.,claude-4-opus-YYYYMMDD). - If fail-over to Claude 3 is ever required, surface an explicit notice and bill appropriately.
---
📌 Requested Action
- Investigate and fix routing so Max 20 accounts reach Claude 4 immediately.
- Confirm model version in UI/API for full transparency.
- Provide credit or refund for any period in which Claude 4 access was not delivered.
- Update status page / release notes with a clear RCA if this affects more users.
---
🖇️ Environment
| Item | Value |
| ------------- | ------------------ |
| Subscription | Claude Code Max 20 |
| Region | Europe |
| First noticed | 2025-07-23 |
---
Thanks for looking into this quickly—happy to supply additional logs or run targeted tests if needed.
8 Comments
Meanwhile, Claude is here to cheer me up 💯
<img width="2518" height="1680" alt="Image" src="https://github.com/user-attachments/assets/ed33f699-4a14-4dca-a454-403f9ee63888" />
explains my experience from yesterday. for me it was showing claude 4 opus but with performance of claude 3.
https://github.com/anthropics/claude-code/issues/4162
@Jo-Baeyaert I'm glad you posted this.
I recently did a Claude trial. I enjoyed the experience. But I was hitting the limits on the trial, so it was clear I would need to pay for max if I became a customer.
But having seen your post today, I'm now having second thoughts.
Did you reach out to Claude directly, i.e. as a customer-to-supplier customer service interaction ? If so, what did they say ? I'm asking because its also important for people like me to know what a company's customer service is like.
@udf2457 I did not reach out to CS for this issue. I can have a look and try to reach out to CS. I'll get back to you on this.
_Claude on my computer keeps rambling as if he has a personal issue with Anthropic 🍿_
Instead of helping me finish my code, he's trying to cheer me up
Well, at least I do feel a bit better, lol 🍭
<img width="1185" height="669" alt="Image" src="https://github.com/user-attachments/assets/c917c7f2-aa8c-4015-b23b-18aba2c437c6" />
This is a strange and unfortunate hallucination on Claude's part. The model you see in
/modelis the model you're actually getting. I'll look into why this might be happening because it's pretty unusual, but I'm certain you were using Opus 4 under the hood. (Try switching to Opus 3 to see how different it would be.)<p>Hi @reverie </p>
<p>Thanks for the prompt response. That said, explanations like “hallucination” or “you were definitely on Opus 4” are insufficient in an enterprise context. This is a contractual-level issue, and we need data—not reassurance.</p>
<h3>1 — Evidence Required</h3>
<p>Please provide the following artefacts for the session observed on <strong>2025‑07‑23 from 07:30 to 09:00 UTC</strong>:</p>
Item | Required Details
-- | --
Inference / Gateway Logs | Timestamped log entries with Request-ID, received model parameter, and the final model actually served.
Canonical Model Identifier | The current, official model-ID string for “Claude Opus 4” (e.g., claude-4-opus-YYYYMMDD) for local verification.
Banner Generation Logic | Source of truth and logic behind the “This codebase was generated by…” banner—where is the model name pulled from?
<h3>2 — Clarifications Needed</h3>
<ol>
<li>
<p><strong>Failure Modes</strong>: What concrete scenarios could result in both the banner <em>and</em> the <code inline="">/model</code> endpoint returning <strong>Claude 3 Opus (20240229)</strong> if <strong>Claude 4</strong> was actually served?</p>
</li>
<li>
<p><strong>Impact Scope</strong>: Was this an isolated incident, tenant-wide issue, or a broader platform-level anomaly?</p>
</li>
<li>
<p><strong>Preventive Measures</strong>: What mechanisms currently exist—and what will be introduced—to ensure model mismatches like this cannot occur again?</p>
</li>
</ol>
<h3>3 — Path Forward</h3>
<ul>
<li>
<p>If logs confirm <strong>Claude 3 Opus</strong> was served: please detail remediation actions (e.g., service credit, retroactive usage correction) and resolution timeline.</p>
</li>
<li>
<p>If <strong>Claude 4 Opus</strong> was served: explain why downstream components reported the wrong model and how this will be prevented going forward.</p>
</li>
</ul>
<p>Thanks in advance for addressing this with the required depth.</p></body></html>
This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.