The Token Company

The Token Company provides a prompt compression API that removes semantic redundancy and context bloat from LLM inputs before they reach your model. Using their bear-1.x model family, developers can reduce token counts by up to 75%, cutting LLM costs dramatically while simultaneously improving accuracy and reducing latency. The API integrates in minutes as drop-in middleware with a single POST call, and benchmarks show measurable improvements on real-world financial documents and reading comprehension tasks.

• bear-1.x Compression Models: Use bear-1, bear-1.1, or bear-1.2 (recommended) to semantically compress prompts while preserving intent and logical relationships.
• Usage-Based Pricing: Pay only $0.05 per 1M tokens removed — you are never charged for tokens that remain in the output.
• One-Call Integration: Send text to POST api.thetokencompany.com/v1/compress with your API key and receive compressed text back; drop it in before any LLM call.
• Adjustable Aggressiveness: Control compression intensity with a aggressiveness parameter from 0.0 to 1.0 to balance compression ratio vs. fidelity.
• Protected Tokens: Wrap sensitive or critical text in <ttc_safe> tags to prevent those sections from being compressed.
• Gzip Support: Enable gzip encoding on requests for up to 2.5x faster large-payload transfers; enabled by default in the Python SDK and npm package.
• Python SDK & npm Package: Get started quickly with official SDKs that handle authentication, gzip, and response parsing out of the box.
• Proven Benchmarks: Compression improved SEC filing QA accuracy by 2.7pp with 20% fewer tokens, SQuAD 2.0 accuracy by 4.0pp with 17% fewer tokens, and reduced E2E latency by up to 37% on Claude Opus.
• Chat & Document Use Cases: Expand conversation history 3x within the same context window, or process large PDFs and web scrapes without bloated inputs.