Assertion Supported AI assessment confidence: 95% certainty 4/5 debate potential 2/5

Ma: Voyage AI Produces Embeddings 3-4x Smaller Than Competitors

Tengyu Ma · No Priors Ep. 67 | With Voyage AI Co-Founder and CEO · Jun 6, 2024 · at 26:39

Tengyu Ma, CEO of Voyage AI, describes how reducing vector dimensions helps optimize search latency in vector databases.

0:00 / 0:06exact quote · 6.8s

▶ Watch the full episode on YouTube → 720p mp4 · rendered on demand · StarZero watermark

“So we produce embeddings that is like a three X, you know, four X smaller dimension than some of the competitors.”

quote is from the automated transcript, cleaned for reading: filler sounds and stutters are removed, nothing is rephrased. names can be misheard (the analysis reads context, assessments check outside sources). how →

The claim, plainly

Voyage AI produces embeddings that have 3x to 4x smaller dimensions than some competitors.

a quote alone can be ambiguous, so this restates it with the references and context of the conversation around it resolved. AI-written; the tape below the player is the source.

Assessment

As of June 2024, Voyage AI's primary models (including voyage-large-2-instruct and voyage-2) produced embeddings with a dimension of 1,024. In comparison, prominent competitor models featured significantly larger default dimensions: OpenAI's text-embedding-3-large had a default dimension of 3,072 (exactly 3x larger, making Voyage 3x smaller), and leading 7B-parameter embedding models on the MTEB benchmark (such as SFR-Embedding-Mistral, NV-Embed-v1, and e5-mistral-7b-instruct) used 4,096 dimensions (exactly 4x larger, making Voyage 4x smaller). Because the claim specifically asserted dimensions 3x to 4x smaller than 'some competitors,' the published specifications fully substantiate the statement.

Voyage AI's voyage-large-2-instruct (released May 2024) and voyage-2 output vector dimensions of 1,024.
OpenAI's text-embedding-3-large (released January 2024) has a default dimension of 3,072 (3x larger than Voyage's 1,024).
Leading MTEB competitor models in mid-2024 based on Mistral/7B architectures (e.g., SFR-Embedding-Mistral, NV-Embed-v1) had native dimensions of 4,096 (4x larger than Voyage's 1,024).

Checked against

[1] voyageai.com [2] zilliz.com [3] voyageai.com [4] medium.com [5] milvus.io [6] thenile.dev [7] grizzlypeaksoftware.com [8] callmissed.com

voyage ai embedding models dimensions 1024 512 256 "voyage-2" dimension "voyage-lite-02-instruct" "voyage ai" models dimensions "june 2024" OR "may 2024" OR "2024" "voyage-2" "voyage-large-2" "voyage-lite-02-instruct" OR "voyage-2" OR "voyage-large-2" dimension openai text embedding 3 large small dimensions cohere embed v3 dimensions google gecko voyage ai dimensions 256 512 1024 1536 3072 "MTEB" leaderboard "voyage-large-2-instruct" OR "voyage-lite-02-instruct" 4096 1024

assessed as of 2026-09-03

This assessment is an AI reading of a public statement, made against the cited references with the confidence shown. It can be wrong. The quote, the tape and the references are all on this page; if something is off, use report an error below. full disclaimer →

Mentioned

Voyage AI

⚑ Spot an error? Report it

More from Tengyu Ma

Prediction Held up

Ma: RAG Will Remain Much Cheaper Than Long-Context Windows

“My prediction is that reg will be much cheaper than long contacts going forward.”

Tengyu Ma Jun 6, 2024 ▶ 14:11 No Priors Ep. 67 | With Voyage AI Co-Founder and CEO

Prediction Not checkable as stated

Ma: AI Is Running Out of Compute and Training Data

“My vision is that in the future the efficiency is very important because we are running out of data and compute. So we have to either use the data much better and use the compute much better.”

Tengyu Ma Jun 6, 2024 ▶ 1:23 No Priors Ep. 67 | With Voyage AI Co-Founder and CEO

Assertion Supported

Ma: Sophia Optimizer Improves LLM Pre-Training Efficiency by 2x

“One of the paper we wrote last year was Sophia which we found, where we found that we have a nutrient optimizer, which can improve the training efficiency by two X for pre-training.”

Tengyu Ma Jun 6, 2024 ▶ 2:59 No Priors Ep. 67 | With Voyage AI Co-Founder and CEO

Assertion Not checkable as stated

Ma: Meta Achieved 1.6x Training Efficiency Gain Using Sophia Optimizer

“And recently I think one of the Facebook friends actually used this in their large scale multimodal training. And they found that in on that scale, I don't know exactly how many parameters there are, but I think, I assume it's kind of more than a hundred billi…”

Tengyu Ma Jun 6, 2024 ▶ 3:54 No Priors Ep. 67 | With Voyage AI Co-Founder and CEO

Insight

Ma: Fine-Tuning Often Fails Due to Data Demands and Hallucinations

“Fine tuning in many cases doesn't work because you need a lot of data to see the results and there are still hallucinations even after fine tuning.”

Tengyu Ma Jun 6, 2024 ▶ 12:06 No Priors Ep. 67 | With Voyage AI Co-Founder and CEO

Opinion

Ma: 100M-Token Enterprises Cannot Afford Long-Context Inference Costs

“So right now, if they have a hundred million tokens, I don't think they can use long context transformers at all because it's way too expensive.”

Tengyu Ma Jun 6, 2024 ▶ 17:43 No Priors Ep. 67 | With Voyage AI Co-Founder and CEO

Ma: Voyage AI Produces Embeddings 3-4x Smaller Than Competitors

More from Tengyu Ma

Turn any episode into a week of clips.