-
[BE] fix: discount cached tokens for Claude on Vertex AI (#7016)
frostbyte_neo 发布于
2026-06-10 13:03:12 +00:00 Adds vertex_ai-anthropic_models to PROVIDERS_CACHE_COST_CALCULATOR so Claude-on-Vertex spans use the Anthropic cache calculator (textGenerationWithCacheCostAnthropic) instead of falling back to textGenerationCost. Without this entry prompt-cache tokens were billed at the full input rate rather than the configured cache-read rate. Follow-up to #6980, which noted this gap in its PR body.
Co-authored-by: Andres Cruz andresc@comet.com
下载附件