Loading startup
Market data is refreshed once per day from public sources. Information may be incomplete or outdated — verify independently before making decisions. This is not investment advice.
Evidence-bound summary — expand sections for movement, risks, and signals.
Memo snapshot · May 20, 2026, 6:12 PM
DealFlow OS uses public web data and automated enrichment. Research may be incomplete, outdated, or incorrect. Verify important information before making investment or outreach decisions.
TL;DR
Seed (YC)The Token Company
Raised $500K across 1 funding round. Latest: $500K Pre-seed (Jan 2026). Investors: Y Combinator. (High).
Funding
Raised $500K across 1 funding round. Latest: $500K Pre-seed (Jan 2026). Investors: Y Combinator. (High).
Hiring
1 hiring‑related row(s); role‑spam risk if mostly generic boards (Low).
Product / news
8 product/news‑styled row(s); headline risk without filings (High).
Verified facts
+5 more in Recent movement below
Winter 2026 (W26) batch. LLM input compression middleware (bear-1 / bear-1.1) cutting tokens ~66% in <100ms.
Weak funding evidence: mention does not include a parsed raise amount/round and may refer to non-funding context.
No open roles indexed yet.
The index price and activity score are algorithmic estimates based on observed public company-level signals. They may be incomplete, stale, or inaccurate and are not investment, legal, tax, or business advice.
Source types found
Strongest / recent news-style rows
Wed, May 20, 06:12 PM · confidence 88%high quality
Pricing · Sun, May 10, 10:48 PM · confidence 90%high quality
wikipedia · Mon, Jun 29, 10:26 AM · confidence 50%medium quality
Newest first · 21 event(s)
Source: Blog
Cut OpenAI, Anthropic, and Gemini API costs with accuracy held flat — or take a smaller cut and lift accuracy by several points instead. The bear-2 prompt compression API strips low-signal tokens from your inputs before they hit the LLM. Works with GPT, Claud…
Source: Homepage
Cut OpenAI, Anthropic, and Gemini API costs with accuracy held flat — or take a smaller cut and lift accuracy by several points instead. The bear-2 prompt compression API strips low-signal tokens from your inputs before they hit the LLM. Works with GPT, Claud…
Source: official_site
The Token Company Jobs You need to enable JavaScript to run this app.
Source: official_site
Prompt Compression API — Cut LLM Token Costs | The Token Company Dark Mode 1280x128 SVG 1280x128 PNG (transparent) 256x256 SVG 256x256 PNG (transparent) 512x512 PNG (transparent) 1024x1024 PNG (transparent) Light Mode 1280x128 SVG 1280x128 PNG (transparent) 2…
Source: official_site
Pricing | The Token Company Dark Mode 1280x128 SVG 1280x128 PNG (transparent) 256x256 SVG 256x256 PNG (transparent) 512x512 PNG (transparent) 1024x1024 PNG (transparent) Light Mode 1280x128 SVG 1280x128 PNG (transparent) 256x256 SVG 256x256 PNG (transparent)…
Source: Blog / news
Pax Historia, processing 193B tokens/month on OpenRouter, ran a 268K-vote model arena with bear-1.1 compression. Compressed models scored higher and A/B tests showed +5% purchase amount lift.
Source: Blog / news
Helonic runs AI agents on construction drawings at near million-token prompts. bear-1.2 compression trims tokens while preserving every critical detail.
Source: Blog / news
Bear-2 compression improved CoQA accuracy from 93.3% to 95.3% while cutting tokens by 8.2%. Removing filler helps the model focus.
Source: Blog / news
Bear-2-Safety compresses input to safety classifiers by up to 30% while preserving or improving F1 across Llama-Guard, ShieldGemma, and Gemini. Unsafe content is preserved at 95-100% retention.
Winter 2026 (W26) batch. LLM input compression middleware (bear-1 / bear-1.1) cutting tokens ~66% in <100ms.
Source: Careers
Join The Token Company. We're hiring ML engineers, software engineers, and more to build the compression layer for LLMs.
Source: Blog / news
Most RAG applications over-fetch context by 3-5x. Break down where your retrieval tokens go and how compression, reranking, and chunk hygiene cut costs.
Source: Blog / news
A practical framework for reducing LLM API costs across system prompts, conversation history, RAG context, tool schemas, and output. Real pricing math included.
Source: Blog / news
bear-1.1 is the latest LLM input compression model from The Token Company. An improved version of bear-1 with better accuracy preservation and faster compression speeds. Reduce AI costs by 3x.
Source: Blog / news
bear-1 is an LLM input compression model by The Token Company that reduces tokens by 66% while maintaining or improving accuracy. Reduce AI costs by 3x with a single API call.
Source: Pricing
Simple pricing for LLM input compression. Free tier to get started, Pro for production, Enterprise for scale.
Source: Careers page
The Token Company Jobs
Source: npm_registry
Node.js SDK for The Token Company — compress LLM prompts to reduce costs and latency
Compresses 100k tokens in under 100ms.
4 row(s)
The company's own site — the authoritative description of what they sell and to whom. Marketing-controlled, so treat claims as positioning rather than verified traction.
Cut OpenAI, Anthropic, and Gemini API costs with accuracy held flat — or take a smaller cut and lift accuracy by several points instead. The bear-2 prompt compression API strips low-signal tokens from your inputs before they hit the LLM. Works with GPT, Claud…
Why it matters: Primary source — the company's own positioning; best read for what they sell and to whom, not for traction claims.
Open source ↗The Token Company Jobs You need to enable JavaScript to run this app.
Why it matters: Primary source — the company's own positioning; best read for what they sell and to whom, not for traction claims.
Open source ↗Prompt Compression API — Cut LLM Token Costs | The Token Company Dark Mode 1280x128 SVG 1280x128 PNG (transparent) 256x256 SVG 256x256 PNG (transparent) 512x512 PNG (transparent) 1024x1024 PNG (transparent) Light Mode 1280x128 SVG 1280x128 PNG (transparent) 2…
Why it matters: Primary source — the company's own positioning; best read for what they sell and to whom, not for traction claims.
Open source ↗Pricing | The Token Company Dark Mode 1280x128 SVG 1280x128 PNG (transparent) 256x256 SVG 256x256 PNG (transparent) 512x512 PNG (transparent) 1024x1024 PNG (transparent) Light Mode 1280x128 SVG 1280x128 PNG (transparent) 256x256 SVG 256x256 PNG (transparent)…
Why it matters: Primary source — the company's own positioning; best read for what they sell and to whom, not for traction claims.
Open source ↗4 row(s)
Third-party press coverage. Independent reporting corroborates company claims; repeated coverage across outlets is a momentum signal.
Simple pricing for LLM input compression. Free tier to get started, Pro for production, Enterprise for scale.
Why it matters: Independent coverage — third-party corroboration of company claims; recurring coverage indicates rising visibility.
Open source ↗Why it matters: Independent coverage — third-party corroboration of company claims; recurring coverage indicates rising visibility.
Open source ↗Why it matters: Independent coverage — third-party corroboration of company claims; recurring coverage indicates rising visibility.
Open source ↗Compresses 100k tokens in under 100ms.
Why it matters: Independent coverage — third-party corroboration of company claims; recurring coverage indicates rising visibility.
Open source ↗1 row(s)
Funding announcements and investor-database records. The strongest public signal of capitalization: round, amount, and syndicate quality when disclosed.
Winter 2026 (W26) batch. LLM input compression middleware (bear-1 / bear-1.1) cutting tokens ~66% in <100ms.
Why it matters: Funding signal — Pre-Seed per this source; verify against the linked original before relying on it.
Open source ↗2 row(s)
Open roles and careers pages. Active hiring implies runway to spend and shows where the company is investing (engineering vs GTM vs ops).
Join The Token Company. We're hiring ML engineers, software engineers, and more to build the compression layer for LLMs.
Why it matters: Hiring signal — open roles imply runway to spend and show where the company is investing.
Open source ↗The Token Company Jobs
Why it matters: Hiring signal — open roles imply runway to spend and show where the company is investing.
Open source ↗1 row(s)
Public engineering activity. Sustained commits, releases, and stars indicate real product development and, for dev tools, developer adoption.
Node.js SDK for The Token Company — compress LLM prompts to reduce costs and latency
Why it matters: Engineering signal — public repo activity evidences active development and possible developer adoption.
Open source ↗9 row(s)
Company blog and newsletters. Shipping cadence and technical depth of posts hint at product velocity and team quality.
Cut OpenAI, Anthropic, and Gemini API costs with accuracy held flat — or take a smaller cut and lift accuracy by several points instead. The bear-2 prompt compression API strips low-signal tokens from your inputs before they hit the LLM. Works with GPT, Claud…
Why it matters: Company publishing — post cadence and depth hint at product velocity.
Open source ↗Pax Historia, processing 193B tokens/month on OpenRouter, ran a 268K-vote model arena with bear-1.1 compression. Compressed models scored higher and A/B tests showed +5% purchase amount lift.
Why it matters: Company publishing — post cadence and depth hint at product velocity.
Open source ↗Helonic runs AI agents on construction drawings at near million-token prompts. bear-1.2 compression trims tokens while preserving every critical detail.
Why it matters: Company publishing — post cadence and depth hint at product velocity.
Open source ↗Bear-2 compression improved CoQA accuracy from 93.3% to 95.3% while cutting tokens by 8.2%. Removing filler helps the model focus.
Why it matters: Company publishing — post cadence and depth hint at product velocity.
Open source ↗Bear-2-Safety compresses input to safety classifiers by up to 30% while preserving or improving F1 across Llama-Guard, ShieldGemma, and Gemini. Unsafe content is preserved at 95-100% retention.
Why it matters: Company publishing — post cadence and depth hint at product velocity.
Open source ↗Most RAG applications over-fetch context by 3-5x. Break down where your retrieval tokens go and how compression, reranking, and chunk hygiene cut costs.
Why it matters: Company publishing — post cadence and depth hint at product velocity.
Open source ↗A practical framework for reducing LLM API costs across system prompts, conversation history, RAG context, tool schemas, and output. Real pricing math included.
Why it matters: Company publishing — post cadence and depth hint at product velocity.
Open source ↗bear-1.1 is the latest LLM input compression model from The Token Company. An improved version of bear-1 with better accuracy preservation and faster compression speeds. Reduce AI costs by 3x.
Why it matters: Company publishing — post cadence and depth hint at product velocity.
Open source ↗bear-1 is an LLM input compression model by The Token Company that reduces tokens by 66% while maintaining or improving accuracy. Reduce AI costs by 3x with a single API call.
Why it matters: Company publishing — post cadence and depth hint at product velocity.
Open source ↗Sign in as an active team member to view private notes, watchlist controls, transcript evidence, and interaction history.