Database Schema
PostgreSQL tables for usage logging, teams, and budgets.
Toolkit
Interactive tools to test TokenOps optimization strategies. Try prompt compression and model cost comparison live.
Prompt Compressor
Paste a system prompt or query to see how compression reduces token count without losing meaning.
~74 tokens (295 chars)
Click "Compress" to see the optimized version of your prompt.
Model Cost Comparator
Compare monthly costs across all major models for your workload parameters.
ModelInput rateOutput rateCost/requestMonthly costvs. cheapest
🏆 GPT-5.6 Luna$0.2/1M$1.2/1M$0.001000$3,0001.0x
Gemini 3.1 Flash-Lite$0.3/1M$2.5/1M$0.001850$5,5501.9x
Gemini 3.8 Flash$0.75/1M$3.75/1M$0.003375$10,1253.4x
Claude Haiku 4.5$1/1M$5/1M$0.004500$13,5004.5x
Mistral Large$2/1M$6/1M$0.007000$21,0007.0x
Claude Sonnet 5$2/1M$10/1M$0.009000$27,0009.0x
GPT-5.6 Terra$2/1M$12/1M$0.010000$30,00010.0x
Gemini 3.1 Pro$2/1M$12/1M$0.010000$30,00010.0x
GPT-5.6 Sol$4/1M$20/1M$0.018000$54,00018.0x
Claude Opus 5$5/1M$25/1M$0.022500$67,50022.5x
GPT-6 Astra$10/1M$50/1M$0.045000$135,00045.0x
Claude Fable 5.1$10/1M$50/1M$0.045000$135,00045.0x
Routing from Claude Fable 5.1 to GPT-5.6 Luna would save $132,000/month (98% reduction) at 100,000 daily requests.
Reference Implementation
Download starter source files for building your own TokenOps infrastructure.
Budget Guardrails
YAML config for token budget enforcement with alerts.
Tagging Schema
LLM gateway configuration with metadata tagging.