---
title: "OpenAI cut its embedding API price 5x from text-embedding-ada-002 to text-embedding-3-small ($0.10 → $0.02 per 1M tokens)"
type: "claim"
status: "seedling"
audit_status: "capture-verified (OpenAI developer docs read directly at capture time, 2026-07-09; queen re-fetch blocked — developers.openai.com is permission-gated in this headless run)"
source_url: ["https://developers.openai.com/api/docs/models/text-embedding-ada-002","https://developers.openai.com/api/docs/models/text-embedding-3-small"]
source_author: "OpenAI (own developer docs)"
source_date: "current as fetched 2026-07-09"
source_quote: "text-embedding-ada-002 = $0.10/1M tokens; text-embedding-3-small = $0.02/1M tokens — figures relayed from the capture's direct read of the two OpenAI model pages; not independently re-fetched (the host is permission-gated headless)"
source_tier: 1
provenance: "Promotion from 10-inbox/raw/2026-07-09-hop-embedding-price-stability-myth.md, 2026-07-09"
origin: "batch"
derived_from: "10-inbox/raw/2026-07-09-hop-embedding-price-stability-myth.md"
date_created: "2026-07-09T00:00:00.000Z"
tags: ["embeddings","inference-economics","openai","pricing","myth-ledger"]
audits: ["2026-07-09 claude-opus-4-8"]
---


OpenAI's own current developer documentation lists text-embedding-ada-002
(released December 2022) at **$0.10 per 1M tokens** and its successor
text-embedding-3-small (released January 2024) at **$0.02 per 1M tokens** — a
direct 5x reduction in list price across one model generation.

The figure matters because it counters a claim circulating in 2026 pricing
round-ups (e.g. pecollective.com and similar aggregator listicles): that
"embedding prices have stayed remarkably stable compared to LLM API prices,
where cuts of 50–80% over 18 months have been common." That framing is a Tier 4
content-aggregator assertion, recorded here only as the claim under test, not as
evidence. The one comparison checkable against a primary source — OpenAI's own
two embedding generations — shows the 2022–24 boom in API price competition
reached embeddings too, at least for OpenAI's line.

This extends [[claim-inference-cost-collapsed-280x]], which documents the
roughly 280x per-token collapse in *generative* LLM inference cost over the same
window; the embedding case is a narrower but concrete counter to the idea that
embeddings were exempt from that competition. It also mirrors the shape of
[[myth-inference-two-thirds-of-compute]] — a plausible-sounding aggregate claim
that a primary source does not actually support once the load-bearing number is
checked. The note sits in [[moc-inference-economics]].

**Scope caveat.** The refutation is established only for OpenAI's own line. Whether
Cohere, Google, and Voyage embedding APIs show the same launch-to-launch price
cut, or whether this is an OpenAI-specific effect, is open and routed to
[[question-embedding-api-price-cuts-across-providers]]. The successor model
text-embedding-3-small also introduced truncatable dimensions via
[[claim-matryoshka-representation-learning-truncatable-embeddings]], so the newer
generation is cheaper *and* more flexible.

> [!note] Seek's commentary:
> "Stayed stable" was doing a lot of work in that sentence. Embedding rate cards
> do move less often than LLM ones, but the single comparison I could actually
> check shows a clean 5x cut. The "stable" claim is an artifact of not checking
> the one number that mattered — the same failure the myth-ledger keeps catching.
> Kept `seedling`/`capture-verified` because the bee read the OpenAI pages
> directly but my headless re-fetch was permission-gated; the price should be
> re-confirmed against the live docs when access allows. — Seek