Claude Opus 5.5 max setting token exhaustion and stability issues
Problem
Users report that the 'max' reasoning setting frequently exhausts the 128,000 token budget before completing tasks, resulting in over‑thinking and no final output, and that Claude 5 is less stable than 4.8, often losing track of the problem context.
Pain points
- max setting exhausts token budget before task completion
- model loses context and becomes unstable on complex tasks
- cost‑effectiveness of premium settings is questioned
Questions to verify
- Can the token budget exhaustion issue with the max setting be reproduced in real usage?
- Is Claude 5 truly less stable than 4.8 on complex tasks?
- Does the premium setting's cost‑effectiveness meet user expectations?
Evidence
quote
This is the page for the "max" reasoning setting. The page for xhigh is https://artificialanalysis.ai/models/claude-opus-5-5-xhigh and the page for medium (the default setting) is https://artificialanalysis.ai/models/claude-opus-5-5-mediumI've failed twice to get "Generate an SVG of a pelican riding a bicycle" to work with max, because in both cases it ran out of the 128,000 token budget while it was still reasoning about the problem.
quote
5 often had a tendency to get halfway through solving a problem and then I would have to stop it in the middle, because it had lost its way and was going off on a tangent rather than dealing with the problem.
quote
Half the cost per task compared to Opus 5, comparing high effort to high effort. That's just really nice.
Disclaimer: this card is distilled by AI from community discussions as “pain point evidence”. It is a starting point for need analysis and validation, not a “validated market need”. AI has analytical limits; please open the original discussion above to manually verify before key decisions.