Analysis of GPT-5.5 Codex Shows Token Clustering May Affect Performance

Jul 4, 2026

AI Summary

An analysis of GPT-5.5 Codex responses reveals a significant clustering of outputs at specific reasoning token counts, particularly at 516 tokens. This clustering is associated with lower reasoning-token intensity and may contribute to performance issues in complex tasks.

GPT-5.5 responses show a disproportionate clustering at 516 reasoning tokens, with additional spikes at 1034 and 1552 tokens.
This pattern is model-specific and correlates with a decrease in overall reasoning-token intensity, potentially explaining degraded performance on complex tasks.
From February to June, the mean and P90 reasoning-token intensity dropped, while the exact-516 clustering increased sharply.
GPT-5.5 accounts for 19.3% of responses but 82.0% of exact-516 events, indicating a significant anomaly compared to other models.
The clustering at fixed values suggests a potential underlying issue with reasoning-budget behavior or internal thresholds in the model.
Further investigation is suggested to determine if the clustering is expected behavior or indicative of a problem with the model's response generation process.

gpt-5.5codexperformancereasoningclustering

Analysis of GPT-5.5 Codex Shows Token Clustering May Affect Performance

Related Stories

OpenAI says GPT 5.6 is the ‘preferred model’ for Microsoft Copilot amid breakup chatter

OpenAI introduces GPT-5.6 models with enhanced efficiency and cybersecurity features

OpenAI Releases GPT-5.6 Model with Enhanced Capabilities

OpenAI launches GPT-5.6 Sol, improving token efficiency for coding tasks