Secondary AI inference markets traded $7.1M in tokens at 86.6% discount to lab list prices.
AI & Agents ·
Secondary AI inference markets cleared $7.1 million in token liquidity over a 24-hour period at an average discount of 86.6% relative to official pricing from frontier labs, according to market data. The activity spans 371 models across multiple routing platforms, with 29,115 active seller offers and 21 billion tokens transacted. Specific examples show Claude Opus 4.8 trading at $1.14 per million tokens versus a $6.00 lab list price, and Kimi K3 at $0.18 per million tokens against a $3.00 official rate.
The secondary markets aggregate pricing across multiple inference providers including Venice, Bedrock, Jatevo, and OpenRouter, enabling dynamic routing to the lowest-cost healthy provider. Claude Opus variants and Kimi K3 dominate volume, with the latter absorbing 4.8 billion input tokens daily and Opus models capturing over 70% of settled dollar volume. Buyers are realizing 75% to 84% execution savings on actual fills compared to direct lab pricing.
The structural question remains whether these secondary market discounts will force official lab price reductions or whether the gap reflects different service tiers, latency profiles, or reliability guarantees. The scale of idle generative media GPU liquidity—over $4.43 million trapped across image and video clusters—and the sustainability of secondary market margins below 10% for commodity open-weight models are also unresolved.