> gpt-4.1: gpt-5 with minimal or low reasoning is a strong alternative. Start with minimal and tune your prompts; increase to low if you need better performance.
Interesting, implies that they still have not fully solved the tool call issues "reasoning" often introduces. Neither Google, nor Anthropic or OpenAI seem to have a reliable solution for that yet. Also, GPT-5 has only 40% of GPT-4.1s context window, so not a full drop in replacement for all use cases, though I have never been able to reliably test any model beyond 250k [0], so potentially this won't be an issue for most.
[0] Not because of any issue with Gemini 2.5 Pro or GPT-4.1, but because I personally have a hard time coming up with realistic test scenarios of such massive size. Just dumping a large code base easily can fill even a 1m context window, but testing whether everything gets properly parsed is challenging. Exception is Llama 4, which has issues even at 75k, despite the 10m context window technically supported.
> gpt-4.1: gpt-5 with minimal or low reasoning is a strong alternative. Start with minimal and tune your prompts; increase to low if you need better performance.
Interesting, implies that they still have not fully solved the tool call issues "reasoning" often introduces. Neither Google, nor Anthropic or OpenAI seem to have a reliable solution for that yet. Also, GPT-5 has only 40% of GPT-4.1s context window, so not a full drop in replacement for all use cases, though I have never been able to reliably test any model beyond 250k [0], so potentially this won't be an issue for most.
[0] Not because of any issue with Gemini 2.5 Pro or GPT-4.1, but because I personally have a hard time coming up with realistic test scenarios of such massive size. Just dumping a large code base easily can fill even a 1m context window, but testing whether everything gets properly parsed is challenging. Exception is Llama 4, which has issues even at 75k, despite the 10m context window technically supported.