4 ms·
Now that the dust is settling a little bit, I have published my notes so far on o3-mini here: https://simonwillison.net/2025/Jan/31/o3-mini/ https://simonwillis
by simonw 2y ago
Now that the dust is settling a little bit, I have published my notes so far on o3-mini here: https://simonwillison.net/2025/Jan/31/o3-mini/ https://simonwillison.net/2025/Jan/31/o3-mini/
To save you the click: I think the most interesting things about this model are the price - less than half that of GPT-4o while being better for many things, most notably code - and the increased length limits.
200,000 tokens input and 100,000 output (compared to 128k/16k for GPT-4o and just 8k for DeepSeek R2 and Claude 3.5 on output) could open up some interesting new applications, especially at that low price.
- nprateem 2y agoI've found with sonnet that the longer its output the less it adheres to the prompt. Ask it to write something and not sound robotic and it's exceptional up to 1000 words. Then it starts becoming more and more robotic. At 3500 words every last paragraph starts "Remember,...", etc