Some big news landed this week around AI-assisted coding!

Cursor launched Composer, a new agent model that achieves frontier coding results while being 4x faster than similar models.

Cognition launched SWE-1.5, another agent model achieving near-SOTA performance while being capable of serving up to 950 tokens/second—that’s really fast! Cognition partnered with Cerebras Systems, the king when it comes to inference speed!

For reference, Sonnet 4.5 — which is widely regarded as the best coding model — achieves a 43.6% score in the SWE-Bench Pro benchmark. SWE-1.5 achieves 40%, but here’s the kicker: It does so 13x more quickly!

While it’s true that at that speed you want to be right rather than quick, it’s also true that these advancements allow you to try out many different paths at the same time and select the most appropriate!

Links in the comments!