Nathan Lambert on measuring open model inference share: exponential growth vs Google scale
Interconnects (Nathan Lambert) · rss · 2026-11-05
Nathan Lambert published research notes on how well we can measure open model inference share, arguing all current metrics are imprecise: noisy platform stats, marketing-flavored startup numbers, imperfect public evaluations. His free Interconnects Dashboard aggregates open model adoption and performance data across the web.
Key points: he uses Baseten, Fireworks and Together AI as bellwethers plus OpenRouter's reported open-model token share, since hyperscalers don't break out usage by model category. Token volume is growing near-exponentially, before any GLM 5.2-driven uptick in open agents. Fireworks announced at its first conference that it serves 40T+ tokens daily to customized models, a vertical that grew 2x since summer — total volume likely well above 100T tokens/day this fall. He overlays Google's earnings-disclosed platform token volumes as a grounding anchor to ask whether open inference is growing or a rounding error vs OpenAI/Anthropic. The notes will be mostly paid, with discounts for policy researchers.
More from Venture
- Email AI Sol raises $4M to track 'open loops', not just draft replies — ayushtweetshere · 2026-11-05
- Perplexity usage up 5x YoY as agents need search — your next buyer is an agent — ramagetime · 2026-11-05
- Of 370 deals reviewed, 66% were passed on due to market competition — dunkhippo33 · 2026-11-05
- October deal flow breakdown: AI/ML founders make up 41% of applications — dunkhippo33 · 2026-11-05
- A fund got 2,718 deals in October and published its full pass stats and reasons — dunkhippo33 · 2026-11-05
- Nace AI offers 120% cashback to switch teams to Drex Max, claims 11x cost edge over Claude Code — testingcatalog · 2026-11-05