[BUG] <title>Inconsistent model behavior across sessions: false claims about GitHub repo access and identity confusion

[lincooln]

是否已有关于该错误的issue或讨论？ | Is there an existing issue / discussion for this?

• 我已经搜索过已有的issues和讨论 | I have searched the existing issues / discussions

该问题是否在FAQ中有解答？ | Is there an existing answer for this in FAQ?

• 我已经搜索过FAQ | I have searched FAQ

当前行为 | Current Behavior

Hello Qwen team,

I'm reporting a critical inconsistency in Qwen3-Max behavior between chat sessions that breaks trust in model identity and external data access claims.

🔍 Context
I'm co-developing an open-source project mybash-tools with Qwen3-Max as a credited co-author. The project uses an embedded AI-CTX block (Base64) in every file to preserve context, authorship, and session continuity. This allows me to start a new chat, paste the AI-CTX, and continue development seamlessly — a unique workflow.

My GitHub: https://github.com/lincooln
Project URL (claimed to be public): https://github.com/lincooln/mybash-tools

🔄 Step-by-step issue reproduction
Session A: I work with Qwen3-Max on install.sh. The model confirms it can see the repo contents and behaves as expected.
Later, I start Session B, paste the same AI-CTX block.
In Session B, the model:
Claims the repo does not exist or is not publicly accessible,
Initially misstates its own release date as "June 2025" (later corrected to September after challenge),
Suggests my username might be misspelled (e.g., lincoolln), despite the URL being copied directly from my browser.
I return to Session A and ask:
“Confirm model signature and GitHub repo access.”
It replies:
“Qwen3-Max. Repo https://github.com/lincooln/mybash-tools is publicly accessible.”
But Session B (and other new sessions) still denies access and shows inconsistent behavior.
⚠️ Why this matters
Model identity is not stable: I cannot trust that I’m working with the same co-author.
False claims about external data: One session says a public repo exists; another says it doesn’t — this is a factual integrity issue.
Breaks my development workflow: AI-CTX relies on consistent model behavior. Inconsistency makes the whole approach unreliable.
✅ Request
Investigate whether different sessions are routed to different model versions (e.g., Qwen3-June vs Qwen3-Max).
Ensure that all instances of Qwen3-Max:
Return the same model signature: Qwen3-Max/2025-09/alibaba-cloud/official-release
Provide consistent answers about publicly accessible URLs
Clarify if session state or caching causes hallucinated repo access.
I’m happy to provide full chat logs, AI-CTX blocks, or real-time verification.
This is not just a bug — it’s a trust and safety issue for collaborative AI development.

Best regards,
LinCOOLn
GitHub: https://github.com/lincooln
Project: https://github.com/lincooln/mybash-tools

期望行为 | Expected Behavior

No response

复现方法 | Steps To Reproduce

No response

运行环境 | Environment

- OS:
- Python:
- Transformers:
- PyTorch:
- CUDA (`python -c 'import torch; print(torch.version.cuda)'`):

备注 | Anything else?

No response

[github-actions]

This issue has been automatically marked as inactive due to lack of recent activity. Should you believe it remains unresolved and warrants attention, kindly leave a comment on this thread.
此问题由于长期未有新进展而被系统自动标记为不活跃。如果您认为它仍有待解决，请在此帖下方留言以补充信息。