Fireworks serves 40T tokens a day, OpenRouter shows just 40-80B — why OpenRouter isn't the whole inference market
AccBalanced · x · 2026-10-26
aicanaries challenges the habit of treating OpenRouter stats as the full inference market: in July Fireworks AI announced serving 40T tokens per day (alongside its Series D and $1B ARR), while OpenRouter showed only 40-80B. In August Together AI was reported at 400T tokens monthly (13.3T daily) vs OpenRouter's 65B. Nebius was reportedly already processing 200B daily for a single client back in Nov '25.
The author argues OpenRouter is far from the endgame: no Claude Code moment, limited open-source spread, fewer capabilities, higher cost. A quoted reply adds that after Nebius Token Factory enabled prefix caching, the resulting stat changes were misread as weak inference revenue — even though Fireworks, Together, Baseten, and Modal have had prefix caching for over a year.
More from Infra
- Anthropic engineer: Claude Code cache hit rates are now high, savings live — teortaxesTex · 2026-10-26
- Six-month bet: hybrid AI deployments, not a wholesale shift to local models — yi111 · 2026-10-26
- Deco decouples 'thinking' from 'spelling' in generative recommenders for 3.2-8.6x throughput — _reachsumit · 2026-10-26
- Redditor with $4,000-$5,000 budget seeks local AI video generation hardware advice — dewsphere · 2026-10-26
- ThousandML Open-Sources CQ 4-bit QLoRA: 17.4% Lower Val Loss Than Unsloth NF4 — textclf · 2026-10-26
- Why OpenAI and Anthropic Cut LLM Prices by Up to 80% in Months — Azer_pubes33 · 2026-10-26