Fireworks AI Releases Fireworks Nexus: A Drop-In Routing and Cost-Control Layer That Moves Routine Coding Work to Open-Weight Models

Fireworks AI released Fireworks Nexus, an AI management and routing platform aimed at engineering organizations. It connects existing coding tools to a managed layer of open-weight models, addressing cost inefficiencies where routine work is run on expensive frontier models. Nexus consists of three components: enterprise controls and cost observability, workflow continuity via FireConnect (a one-line install under Apache 2.0), and intelligent traffic management using a custom model that scores request difficulty. The router is a research preview, currently routing between Claude Opus 5 and GLM-5.2 with an Anthropic key pass-through, or Kimi K3 and GLM-5.2 in all-open mode. Preliminary testing with Notion and Doximity showed a 33% reduction in cost per merged pull request. Faros AI evaluated 211 real engineering tasks across seven routes, finding Claude Code on GLM-5.2 scored 0.568 against 0.521 for Claude Code on Opus 4.8, with costs of $0.92 vs $1.76 per task. Arize evaluated 10 models on 40 Terminal-Bench tasks with 2,400 runs, showing that a deliberate escalation ladder beat every single-model strategy at $0.525 per successful task with 32.3/40 tasks solved, versus $0.636 and 25/40 for GPT-5.5 alone. Three deployment paths exist: FireConnect plugin, direct API configuration, or the router sitting in front of a frontier contract.