Evo2-flash: faster inference engine for the Evo 2 genome language model (Arc Institute and Stanford, Apache-2.0)

Evo2-flash runs Evo 2 (StripedHyena2) on vLLM as a drop-in replacement for the evo2 Python class, built so that long-context in-context-learning prompts are practical. Arc says it generates sequences up to 35x faster and runs on GPUs available in Google Colab. It came out with the Evo 2 in-context-learning preprints of 29 October 2026, which include in-silico bacteriophage genome design (computational, not yet experimentally tested). Repository is Apache-2.0.