GPT-5.5 Choked by '516' Token Limit: 80% of Complex Reasoning Silently Truncated
新智元 · wechat · 2026-07-05
New Smart Era reports that OpenAI's flagship model, GPT-5.5, is exhibiting severe anomalies in complex coding tasks: a massive amount of reasoning tokens are stopping precisely at 516, with similar spikes concentrated at the 1034 and 1552 marks. A statistical analysis covering 390,000 response records revealed that GPT-5.5 accounts for 82% of all "precise 516" events network-wide. In contrast, the overall reasoning intensity during the same period plummeted, creating a contradictory comparison and sparking strong skepticism among developers about whether OpenAI is secretly limiting compute budgets.
A large number of Codex developers have flocked to GitHub Issue #30364 to complain and submit data, demanding an official response from OpenAI: Is this an inference budget limit, a routing issue, or a triggered fallback downgrade mechanism? Although the original poster noted there is no direct evidence yet that OpenAI is actively truncating the chain of thought, the high concentration of anomalous data has attracted widespread attention, prompting some users to start switching back and forth between Codex and Claude.
Related event: GPT-5.5 Reportedly Degrades Due to Silent Token Truncation(3 posts)→
More from Models
- Meta's Muse Agent has built-in invite code logic, hinting at free-usage expansion — testingcatalog · 2026-09-11
- Same Echo Maze prompt, three frontier models: all passed visually but shipped the same hidden bug — eyishazyer · 2026-09-11
- Benchmark scores drop from 89% to 19% on new evals — how benchmaxxing breaks leaderboard trust — airesearch12 · 2026-09-11
- ChatGPT tells user their question is too hard and to 'accept dumber answers' — phido3000 · 2026-09-11
- Claude is no longer available for minors as Anthropic rolls out age assurance — Muhammad523 · 2026-09-11
- Developer Building a Unified Leaderboard of All Model Benchmark Scores — airesearch12 · 2026-09-11