Cody Blakeney

shocking: ai researchers write great papers using pytorch/jax but ask them to write their optimization loops in fortran and suddenly they collapse

We've evaluated a lot of base models on perplexity-based evals and Kimi k2.5 proved to be the strongest! After that, we do continued pre-training and high-compute RL (a 4x scale-up). The combination of the strong base, CPT and RL, and Fireworks' inference and RL samplers make Composer-2 frontier level. It was a miss to not mention the Kimi base in our blog from the start. We'll fix that for the next model.

the model is the product

Your evals, your environments, your data, your model. You need to scale the ladder Using foundation models to build a good harness and getting eval coverage is a great way to start building business value. Building environments to improve on your use cases is something that is often something your domain experts at your company are uniquely positioned to do. CPT on open models once you have done all this leg work is a great, tight feedback loop to insure you have data of sufficient quantity and quality to even consider pretraining. Scale the ladder

Looks like it’s confirmed Cursor’s new model is based on Kimi! It reinforces a couple of things: - open-source keeps being the greatest competition enabler - another validation for chinese open-source that is now the biggest force shaping the global AI stack - the frontier is no longer just about who trains from scratch, but who adapts, fine-tunes, and productizes fastest (seeing the same thing with OpenClaw for example).

For students or people looking to break into careers in AI this exists to be a talent pipeline tool. Making visible and meaningful entries here is probably one of the highest ROI ways to demonstrate your skills and break in without getting a PhD or publishing.

We've evaluated a lot of base models on perplexity-based evals and Kimi k2.5 proved to be the strongest! After that, we do continued pre-training and high-compute RL (a 4x scale-up). The combination of the strong base, CPT and RL, and Fireworks' inference and RL samplers make Composer-2 frontier level. It was a miss to not mention the Kimi base in our blog from the start. We'll fix that for the next model.