How well can we measure open model inference share?
AI-assisted digest, checked against the original source before publishing. How we work
Interconnects' first paid research note attempts to size the open model inference market using inference-startup token volumes — and finds growth, but fuzzy data.
The motivating problem is measurement quality. Existing signals on open model adoption are described as incredibly imprecise: noisy platform metrics, startup figures that skew marketing-related, and imperfect public evaluations, all requiring an informed eye to interpret. Hyperscalers complicate the picture further — Amazon Bedrock and Azure's model garden are said to likely have a major footprint in open model inference, but they have not broken down usage by model category, leaving that share largely unobservable.
To approximate the market, the note leans on the open model inference startups Baseten, Fireworks, and Together AI as bellwethers, since their fortunes correlate directly with open model inference as a category, plus OpenRouter's reported open model share of token volume. Public disclosures of those startups' token volumes over time, plotted on a log scale, show near-exponential growth before an uptick following GLM 5.2's described step change in open agents.
Fireworks offers the freshest datapoint: at its first company conference this week, it announced serving more than 40 trillion tokens per day to customers' customized models. That figure covers just one slice of its inference — a vertical the company said grew more than 2x since summer — leading the author to expect its total to land well above 100 trillion tokens when the next update arrives.
For grounding, the note overlays those startup volumes against Google, which discloses platform and API token volumes in its earnings reports, producing a comparison plot of open model inference platforms against a hyperscaler's reported totals.
What remains open is the central question the note sets out to answer: how large the open model inference market actually is, whether it is growing, and whether it is a rounding error relative to OpenAI and Anthropic. Until hyperscalers break out usage by model category and more startup disclosures land, any figure for open model share stays an approximation rather than a measurement.
Key facts
- • Interconnects launched its first open model economy research note for paid subscribers.
- • Note uses Baseten, Fireworks, and Together AI as inference bellwethers.
- • Fireworks reported 40T tokens per day for customers' customized models.
- • That Fireworks vertical grew over 2x since summer, company said.
- • Interconnects Dashboard data remains free for the foreseeable future.
Why it matters
Builders sizing demand for open model serving now have a rough public yardstick: startup token volumes on a log-scale growth curve, cross-checked against Google's earnings disclosures. Expect the next round of startup updates — particularly Fireworks' total — to be the cleanest signal of whether open inference is scaling into a real category or lagging closed-model providers. Until hyperscalers pu
Spotted an error? Send a correction.
Get the day's AI news in one email
The Daily Epoch · 60-second read · unsubscribe anytime
