Fireworks AI VP: Nexus Router Is Itself a Trained Model, Not Rules
sophiamyang · x · 2026-11-04
Chris Harris, VP of Product Engineering at Fireworks AI, explained the design of Nexus, the company's model router, at a live event.
- A "team of models" beats one superstar: models have distinct personalities and strengths; a diverse bench will matter more than any single top model.
- Nexus is a router that is itself a trained model, routing on fit (matching model quirks to task attributes, beyond just the prompt), cost (preserving cache hits, predicting token count, reasoning effort, and tool calls), and personalization (learning who is asking over time).
- Fireworks' moat is not building its own models but the orchestration layer above them.
More from Infra
- Anthropic touts 15-person chip team using Claude to IPO investors, S-1 hints otherwise — eliebakouch · 2026-11-04
- 8 Coding Agents on One RTX PRO 6000: Measured Throughput for Local Qwen Deployment — AIFlow_ML · 2026-11-04
- Agent37 launches $4.76/month AI agent sandboxes, undercutting Vercel's $249 by 98% — rohanpaul_ai · 2026-11-04
- TensorFold engine 1.2.2 ships ~40% faster prefill on GB10, Metal and HIP gains — HankYeomans · 2026-11-04
- mlx-serve PR runs DeepSeek-V4.1 tensor-parallel across two Macs, 57% faster decode — HankYeomans · 2026-11-04
- PyTorch Day Japan 2026 agenda drops: ExecuTorch porting to IBM AIU edge accelerator — PyTorch · 2026-11-04