How much does Google's Gemini Embedding actually cost?

Standard tier is $0.15 per 1 million input tokens for gemini-embedding-001, confirmed directly against Google's own pricing page. A batch tier is available at $0.075 per 1M tokens for workloads that can tolerate asynchronous processing. Google has also since introduced a newer gemini-embedding-2 model at $0.20 standard / $0.10 batch per 1M tokens, which our calculator does not yet list.

How did the wrong price get into the calculator?

Our embeddings cost calculator keeps its own small, hand-maintained price table rather than pulling from the larger pricing dataset that powers the rest of the site. That larger dataset already had the correct $0.15 figure for Gemini Embedding. The calculator's separate table, labeled internally as a June 2026 reference snapshot, was never synced to it and kept a stale $0.025 number that doesn't match anything Google has published.

How much did the wrong price understate a real job?

Take the calculator's own default inputs: 100,000 chunks at 500 tokens each, 50 million tokens total. At the old $0.025 rate that's quoted as $1.25. At the real $0.15 rate it's $7.50 -- six times higher. Anyone who sized a Gemini Embedding budget off our old number was underestimating the bill by about 83%.

Is Google still a cheap embedding option after the fix?

No, and that's the real damage from the bug -- not just the dollar amount but the ranking. At the wrong $0.025 rate, Google landed as the 3rd-cheapest of the seven models in our comparison table, ahead of Voyage, Mistral, Cohere and OpenAI's larger model. At the real $0.15 rate it's the single most expensive model in that same table, behind all six others.

Google's Gemini Embedding Isn't $0.025 per Million Tokens. It's $0.15.

5 November 2026 · Site Methodology & Corrections · 4 min read

I was cross-checking our 埋め込みコスト計算ツール against the larger pricing dataset that powers the rest of the site today, model by model. Six of the seven rows matched exactly. The seventh -- Google's Gemini Embedding -- was priced at $0.025 per 1 million tokens in the calculator and $0.15 in our own dataset elsewhere on the site. Only one of those numbers is real.

The number, checked today

Google's current pricing page lists gemini-embedding-001 at $0.15 per 1 million input tokens on the standard tier, or $0.075 on the batch tier for workloads that can wait. There's also a newer gemini-embedding-2 model at $0.20 standard / $0.10 batch, which neither our calculator nor our wider dataset lists yet. Nowhere does $0.025 appear on Google's side.

Source	Gemini Embedding price
Our embeddings calculator (before today)	$0.025 / 1M tokens
Our own pricing dataset (unchanged, already correct)	$0.15 / 1M tokens
Google's official pricing page, checked today	$0.15 / 1M tokens (standard)

The calculator's number wasn't an old price that changed later -- it never matched anything Google has published. It's a six-times understatement that sat on a live page for months without anyone running the two numbers side by side.

What that did to a real estimate

The calculator defaults to 100,000 chunks at 500 tokens each, 50 million tokens total. At the old $0.025 rate, embedding that corpus once was quoted as $1.25. At the real $0.15 rate it's $7.50. Anyone who ran that number for a real Gemini Embedding budget was underestimating the bill by about 83% -- not a rounding error, a model of the cost that was off by a factor of six.

The part that actually matters: the ranking flipped

The dollar gap is the smaller problem. The bigger one is what it did to the comparison table the calculator shows next to your result. At the wrong $0.025 rate, Google landed as the 3rd-cheapest of the seven models compared -- ahead of Voyage, Mistral, Cohere, and OpenAI's larger model, looking like a solid budget pick. At the real $0.15 rate it's the single most expensive model in that same table, behind all six others including OpenAI's priciest option.

That's not "a bit more expensive than we said." That's a model moving from "looks like a good deal" to "the worst deal on the page," entirely because of a number nobody re-checked.

Why one page drifted from the other

The embeddings calculator keeps its own small, hand-written price table instead of pulling from the larger pricing dataset the rest of the site already uses -- the comment next to it in the source read "reference Jun 2026." That snapshot was taken once and never synced against the dataset that already had the right figure sitting a few files away. Two sources of truth for the same number is exactly how this kind of drift survives: nothing was wrong on the day it was written, and nothing forced a second look afterward.

What I fixed

Updated the Google row in the 埋め込みコスト計算ツール from $0.025 to $0.15 per 1M tokens, matching both Google's current pricing page and the figure our own wider dataset already had correct. Didn't add the newer gemini-embedding-2 tier yet -- that's a separate addition, not part of this fix, and it'll get its own pass once we've verified it the same way.

埋め込みコスト計算ツール — corrected Gemini Embedding rate, last verified 2026-11-05.
How embeddings pricing works — the fundamentals behind per-token embedding costs.
RAG チャットボットのコスト計算ツール — embedding cost is one line in a bigger RAG budget; see the whole picture.

New to usage-based API pricing in general? Start with the 無料の API コストガイド.

Deploy it yourself: DigitalOcean — $200 の無料クレジット ↗ ・ホスティンガー VPS ↗

Checked directly against Google's AI Studio pricing page and our own internal pricing dataset, 2026-11-05. Reference figures, not live quotes -- confirm current rates before budgeting.