Unverified50% confidencePredictionExact time
Inference prices are falling at approximately 10 times per year or faster, while VM and memory rental costs are rising 25% to 30% per year, and the two curves are crossing.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/10/2026
First Seen
Valid until: 1/8/2027
Sources
Related Entities
Related Claims
UnverifiedOver the past two years, the per-token API call cost of mainstream large model vendors has shown a continuous downward trend.72% similarUnverifiedAs the number of agent turns increases and context accumulates, uncached input costs rise on a quadratic curve—when 10 turns correspond to 1x cost, 40 turns correspond to approximately 16x cost.68% similarUnverified推理成本随上下文长度的增长呈近似二次方上升66% similarUnverifiedThe cost of VMs is higher resource overhead and slower startup speed, but in scenarios with high security requirements this additional overhead is worthwhile.66% similarUnverifiedIn 2024, the price per million input tokens for large model APIs fell approximately 90% compared to 2023, dropping from approximately $30 to $2–3.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/994233API
curl https://kongchang.com/api/v1/knowledge/claims/994233MCP
get_claim(id=994233)