Sonnet 4.5
We've gotten used to vendors changing multiple things at once with their flagship releases, but Anthropic is playing a more conservative game.
Sonnet 4.5 is Sonnet 4, but smarter.
We've gotten used to vendors changing multiple things at once with their flagship releases, but Anthropic is playing a more conservative game. Sonnet 4.5 has the same pricing, the same context limits, and the same janky manual prefix cache as Sonnet 4. But now it's smarter.
Here's what difference that makes.
Sonnet has narrowly reclaimed the flagship model coding crown from GPT-5. (Opus 4.1 also beat GPT-5, but while Sonnet 4.5 is 2x as expensive as GPT-5, which can be a reasonable trade for Sonnet's speed advantage, Opus 4.1 is almost 10x as expensive, so I've left it out of the charts here. Of course you can customize the charts yourself on the live page.)
Interestingly, while Sonnet 4.5's price-per-token is unchanged, it uses those tokens more effectively, resulting in about a 40% improvement in real-world cost-effectiveness--almost closing the gap to Gemini 2.5 Pro.
Finally, Sonnet remains the clear leader in speed, with no real competition in sight; it's even faster than Gemini 2.5 Flash. This is part of what makes Sonnet worth the price premium.
Now that both OpenAI and Anthropic have released their next-gen models, the competitive landscape is firming up again. What can we expect from Gemini 3?
