Embedding Models Compared (OpenAI vs Cohere vs OSS)
Benchmark OpenAI vs Cohere Embed vs open-source BGE/E5: dimensions, cost, and MTEB scores.
Embedding Models Compared (OpenAI vs Cohere vs OSS) is a free AI Agents lesson on CoddyKit — lesson 4 of 4. You can read the complete lesson below for free — then practise it hands-on in the browser with a built-in code editor and a 24/7 AI tutor. It is part of the AI Agents learning path, one of 4 lessons in the course, and your progress syncs across the web and the CoddyKit app.
Why Compare?
OpenAI is not the only game in town. Cohere, Voyage, Google, and many OSS models offer competitive embeddings — sometimes cheaper, more multilingual, or specialized for code.
The right choice depends on use case.
OpenAI text-embedding-3
The default for most teams:
- + Strong English quality
- + Cheap ($0.02 / 1M tokens for small)
- + Up to 3072 dim with Matryoshka
- - Cloud-only
Cohere Embed
Cohere's embed-multilingual-v3 is the strongest multilingual option:
- + Excellent multilingual (100+ languages)
- + Separate query and document modes
- - Slightly more expensive
Voyage Embeddings
Anthropic-recommended; strong specialised models:
- + voyage-3 — top general-purpose
- + voyage-code-3 — best for code
- + voyage-finance / law — domain-tuned
BGE (BAAI General Embedding)
The leading open-source family from BAAI:
- + Free, runs on CPU or small GPU
- + bge-large-en, bge-m3 (multilingual)
- - Slightly behind closed models on MTEB
E5 (Microsoft)
Microsoft's e5-large family. Strong OSS contender, popular in research benchmarks.
Choosing by Benchmark
MTEB (Massive Text Embedding Benchmark) is the standard leaderboard: huggingface.co/spaces/mteb/leaderboard
Filter by your task (retrieval, classification, clustering) and language.
Choosing by Cost
For 1M documents (~500M tokens):
- text-embedding-3-small: ~$10
- text-embedding-3-large: ~$65
- Cohere v3: ~$50
- Local BGE on a GPU: ~$5 (electricity)
Choosing by Privacy
If you cannot send data to OpenAI:
- Self-host BGE / E5 with sentence-transformers
- Use Cohere via AWS Bedrock (data-residency commitments)
- Use OpenAI via Azure (similar)
Choosing by Language
For non-English content:
- Cohere multilingual v3 — broadest support
- bge-m3 — strong OSS multilingual
- OpenAI text-embedding-3 — surprisingly good but English-leaning
Domain-Specific Models
For code: voyage-code-3 or jina-embeddings-v2-base-code
For finance/legal: voyage-finance, voyage-law
For science: SPECTER, SciNCL
Mixing Models
Do NOT mix vectors from different models in the same index — they are not comparable. Re-embed everything with one model if you switch.
A/B Test Two Models
The right way to choose:
- Build a small eval set of (query, expected docs) pairs
- Embed your corpus with both models
- Measure recall@10 on the eval set
- Pick the winner
Mixing Models
Can you mix embeddings from two different models in one index?
Recap
Default to text-embedding-3-small; explore Cohere for multilingual, Voyage for code/domains, BGE for self-hosting. Always benchmark on your data.
Frequently asked questions
Is the “Embedding Models Compared (OpenAI vs Cohere vs OSS)” lesson free?
Yes — the full text of “Embedding Models Compared (OpenAI vs Cohere vs OSS)” is free to read here on the web, and the AI Agents course includes 4 lessons in total. To practise it interactively (a built-in code editor and a 24/7 AI tutor) and unlock the rest of the AI Agents course, upgrade to CoddyKit PRO.
What will I learn in “Embedding Models Compared (OpenAI vs Cohere vs OSS)”?
Benchmark OpenAI vs Cohere Embed vs open-source BGE/E5: dimensions, cost, and MTEB scores. You practise AI Agents with hands-on code you run directly in the browser, and a 24/7 AI tutor answers your questions as you work through the lesson.
Do I need any experience to start AI Agents?
No prior experience is required. AI Agents on CoddyKit is structured for beginners through advanced learners; this is — lesson 4 of 4, so you can start here or from the beginning and move at your own pace.
How long does the “Embedding Models Compared (OpenAI vs Cohere vs OSS)” lesson take?
Most CoddyKit lessons take about 5–10 minutes. Each one is bite-sized and interactive, so you make steady progress and pick up exactly where you left off across the web and the app.
Can I write and run code in this AI Agents lesson?
Yes. Every AI Agents lesson includes a built-in code editor, so you write and run real code right in your browser and get instant AI feedback — no local setup required.
All lessons in this course
- What Embeddings Are (Vector Representations)
- Generating Embeddings with text-embedding-3
- Cosine Similarity for Retrieval
- Embedding Models Compared (OpenAI vs Cohere vs OSS)