Why free model tiers look dead when they aren't
Why free model tiers look dead when they aren't I keep a small probe script that walks a list of OpenAI-compatible endpoints and asks each one the same two questions: what do you have, and will you answer a chat right now. For months I only looked at the first answer. That was the mistake. On 09.10.2026 I finally logged both. Groq answered 200 on openai/gpt-oss-20b and qwen/qwen3.8-27b — catalog…
The article discusses the potential pitfalls of relying on free model tiers in OpenAI-compatible endpoints. The author uses a probe script to test various models and endpoints, but initially only checked the response status code without considering the actual functionality. They found that a 200 status code from the catalog endpoint only confirms authentication and catalog access, not the current availability or quota of the models.
A 410 status code indicates that a model has been retired, while a 429 status code suggests a budget issue, not an outage. The author also highlights the importance of differentiating between account-based free tiers and model-named tiers, as a scanner that only looks for a specific naming convention may overlook working providers.
They emphasize the need to log both the status code and the response body, as the latter can provide valuable information about the cause of any issues. Finally, the author recommends treating each provider's error response individually, rather than making assumptions based on the provider itself, as this can lead to more accurate and nuanced insights into the performance of free model tiers.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
