Testing LLM and sharing my personal experience with them
Will OpenRouter be always OpenAI-Compatible API?
Qwen3 14B vs. GPT-4o Mini
di Mohammad Jawad Barati
Short answer, no. But it is OpenAI compatible for the most part or at least what I have seen. BTW if you are curious about my experiment you can look at this Jupyter Notebook: https://github.com/kasir-barati/smart-novel-beatrice/blob/bcb520b3ca73531eb93c650004090865b9f0f7db/local-setup/notebooks/openrouter_qwen_tts_voices.ipynb I love JN for testing and experimenting LLMs. Let me know your goto tool when testing and checking what is possible.
In this episode I compared and talked about these two models. Which one is faster, which one is cheaper and which one I decided to use. And this is the JupyterNotebook I talked about it in this episode: https://github.com/kasir-barati/smart-novel-beatrice/blob/3ee5d89d91f94e4f4903ede1d2009350f040f4fb/local-setup/notebookes/tts_sanitiation.ipynb This is the link in question for comparing the two models on OpenRouter: https://openrouter.ai/compare/qwen/qwen3-14b/openai/gpt-4o-mini
I was in the middle of experimenting with LLMs to generate audio files for my smart-novel app (although at the moment the LLM-powered part lives in https://github.com/kasir-barati/smart-novel-beatrice). That is why I decided on testing Qwen3-TTS and Gemeni-TTS. Let me know what you think? BTW this is the text I gave both LLMs: [dramatic] The system chimed: And this is the style instruction I gave Qwen3-TTS (note that Gemini-TTS does not accept such a thing): A male narrator with clear and easy to understand voice. Treat things like [dramatic], [proud], etc as emotion tags. But make sure to not mistake them for other text which are not emotion tags. Also you should not read < or > in a text, just pause a few milliseconds before reading them.