Cognition launched SWE-2 Flash on November 4 — a 276-billion-parameter mixture-of-experts coding model that runs at 200 tokens per second, three times faster than its SWE-2 flagship, and costs one-fifth as much. It is available free for paid Devin users through November 30. The benchmarks are mixed, the post-trial pricing is undisclosed, and Cognition is keeping the weights private despite building the model on an Apache-licensed open base. Here is what developers actually need to know. Speed Is Real. Price Is a Ratio. The 3x speed claim holds up architecturally. SWE-2 Flash is a mixture-of-experts model with 276 billion total […]
The post
source & further reading
byteiota.com — original article
Recursive Flint 40B: Run a Coding Agent Locally
Google Open-Sources XLA Serving Engine: Serve JAX Models at 36x Throughput
Hugging Face Transformers MIC: Stop Silent ML Regressions