Fireworks launches cache-aware FireRouter with Opus: coding costs cut 57% at 98% accuracy
Madisonkanna · x · 2026-10-03
Fireworks.ai released FireRouter with Opus, a cache-aware router that automatically routes coding requests between Claude Opus 5.5, GLM 5.3 and GLM 5.3 Flash, now available as a standalone serverless endpoint.
- A month-plus of internal A/B testing on real coding traffic: cost per session fell 57% (±19 ppts), from $15.36 to $6.63
- Graded-turn accuracy of 78.7% vs 80.2% for Opus-only — just 1.5 ppts lower
- The router weighs each model's task fit and factors in prompt-cache invalidation costs when switching models
- Fireworks CEO Lin Qiao says their traffic maintains a 94.2% cache hit rate
A concrete production example of cache-aware routing: treating cache migration cost as part of the routing decision to slash agent coding spend.
More from coding & agent
- Obol ships free self-hosted gateway: per-agent keys, budgets, MCP policy — ToBeContinuedHermit · 2026-10-03
- 'Cursor is cooked' takes it back: the hot take that aged like milk — gabriberton · 2026-10-03
- Anthropic releases free workshop on building collaborative multi-agent workflows — s_mohinii · 2026-10-03
- Jevbox: open-source self-organizing document drive with hierarchical search, no vector DB — rickasaurus · 2026-10-03
- AI will write 748 tests — except they are all the wrong ones — chrisalbon · 2026-10-03
- Reddit user ditches Zellij for Herdr to juggle multiple agent CLI sessions — leebase65 · 2026-10-03