Moonshot released K2.7 Code, the latest in their K2 line of coding models, and it's live on Fireworks Day 0, on serverless and the API.
Fireworks AI has launched its K2.7 Code model, an advancement in their K2 line of coding models, now available on their serverless platform and API. This new model is designed to reduce reasoning token usage in long agent loops, leading to faster generations and lower costs per completed task. K2.7 Code achieves this by producing approximately 30% fewer reasoning tokens than its predecessor, K2.6, while simultaneously improving performance on coding benchmarks. AI
IMPACT This release offers improved efficiency for agentic coding tasks, potentially lowering operational costs for AI developers.