You've hit your session limit · resets 8pm (America/Mexico_City)

Bug Description

I would like to submit a complaint regarding my experience using Fable 5, as the frequent interruptions are causing excessive and unproductive consumption of my token quota.

The primary issue is not simply that the model becomes unavailable, but the cost associated with every model switch. Each time Fable 5 stops responding and the system automatically switches to another model (for example, Opus), a significant number of tokens are consumed solely to reconstruct the conversation context. This token usage produces no actual progress on the work I am trying to accomplish.

For example, yesterday my quota had just been reset and was fully available. I asked Fable 5 to continue working on my project. Before making any meaningful changes, the model became unavailable and the system switched to Opus. That transition consumed additional tokens to recover the conversation context. Immediately afterward, the system initiated a conversation compaction process, which also consumed a substantial number of tokens. However, the compaction never even completed because my quota was exhausted during the process.

As a result, an entire freshly reset quota was effectively consumed by a single request, without either Fable 5 or Opus making any useful changes to my project. All of the token usage was spent on reconstructing context due to repeated interruptions and automatic model switches.

I understand that technical limitations may require model switching under certain circumstances. However, I do not believe users should bear the token cost of internal system operations that are entirely outside their control, especially when those operations produce no meaningful work or output.

From my perspective, a significant portion of my quota is being spent repeatedly rebuilding context instead of performing the tasks I requested. This substantially reduces the value of the subscription and makes it difficult to use the service effectively for software development projects that involve long-running conversations.

I would appreciate it if you could review this behavior and consider implementing a mechanism so that interruptions, automatic model switching, and conversation compaction do not count against the user's quota, or at the very least, that such token usage is not charged when no effective work has been performed on the user's request.

Finally, I want to address one additional concern. While I cannot know the reasons behind this behavior, from a user's perspective the experience can create the impression that frequent interruptions and the resulting context reconstruction consume subscription quotas much faster than expected. I am not suggesting this is intentional, but I believe it is important to recognize that this perception can undermine user confidence. Greater transparency about how token usage is accounted for during model switches and internal processes would go a long way toward addressing these concerns.

Environment Info

• Platform: linux
• Terminal: vscode
• Version: 2.1.203
• Feedback ID: e6bda264-298d-4a2f-89b1-fce8c4c2eeda

Errors

[]