Summary
Per-million-token cost breakdown across all Tikoly-supported models. DeepSeek v4 Flash delivers the lowest cost per token for high-throughput agent tasks.

Summary

Tikoly’s pricing team analyzed per-million-token costs across 15 supported models, factoring input, output, cached input, and reasoning token rates. DeepSeek v4 Flash emerges as the most cost-effective option for high-throughput agent workloads, while premium reasoning models (o1, DeepSeek R1) command higher rates for complex multi-step tasks.

Key Findings

  • Input token range: $0.15/M (DeepSeek v4 Flash) to $15.00/M (o1 with reasoning)
  • Cached input discount: 50% off input rate on all supported models when cache hits occur
  • Reasoning token premium: Thinking models can consume 5-10x more tokens on internal reasoning than visible output
  • Package impact: The Enterprise tier (100M tokens at $30) brings effective per-token cost below $0.30/M for most models

Methodology

Each model was benchmarked against five standard task profiles: code generation, translation, PR analysis, multi-turn chat, and deep research. Token counts were captured from pi-ai’s Usage interface and multiplied against tikoly.com’s rate card. The spectrum above shows relative cost intensity across model tiers.