Pricing Comparison

Model	Provider	Dimensions	Price / 1M tokens
`text-embedding-3-small`	OpenAI	1536	$0.02
`text-embedding-3-large`	OpenAI	3072	$0.13
`text-embedding-ada-002`	OpenAI	1536	$0.10
`gemini-embedding-001`	Google	768	Free (under limits)
`gemini-embedding-2`	Google	768	Free (under limits)

Google's Gemini embedding models are free under generous rate limits, making them an attractive alternative for teams looking to reduce embedding costs. OpenAI's models offer higher dimensions and established benchmark performance, but at a per-token cost.

Quality Comparison

Both providers produce high-quality embeddings for retrieval tasks. Key differences:

OpenAI text-embedding-3-large — highest dimensional output (3072d), strong on MTEB benchmarks
Gemini embedding-2 — 768d, optimized for Google's ecosystem, competitive retrieval quality
Higher dimensions don't always mean better quality — task-specific evaluation matters

The Migration Problem

Switching from OpenAI to Gemini (or vice versa) isn't just an API call change. Vectors from different models live in different mathematical spaces. Even if two models output the same number of dimensions, their vectors are incompatible.

Traditionally, migration means re-embedding your entire corpus — which requires:

Access to all original text (which you may not have stored)
Paying for re-embedding millions of documents
Downtime while vectors are being rebuilt
Risk of quality degradation during the transition

How Schift Solves This

Schift uses learned projection matrices to map vectors from one model's space to another's — achieving 99.7% retrieval recovery without re-embedding a single document.

from schift import Schift
s = Schift(api_key="sk-...")

# Migrate your entire vector DB from OpenAI to Gemini
s.upgrade(
  db="my_pgvector",
  to="google/gemini-embedding-2"
)
# → 1M vectors migrated in 2.3s
# → Recovery: 99.7% | Verdict: SAFE

Cost Savings: OpenAI → Gemini

Scenario	OpenAI Cost	Gemini Cost	Savings
1M tokens/month (3-small)	$0.02/mo	$0.00	100%
100M tokens/month (3-small)	$2.00/mo	$0.00	100%
1B tokens/month (3-large)	$130.00/mo	$0.00	100%
Via Schift (5% routing fee)	—	$0.00 + $0	~95%+

With Gemini's free tier and Schift's projection-based migration, teams can eliminate embedding costs entirely for most workloads without quality loss.

When to Stay on OpenAI

You need 3072-dimensional embeddings for maximum precision
Your benchmarks show significantly better performance on your specific domain
You're already at scale and the cost difference is negligible

When to Switch to Gemini

Cost is a factor — Gemini embeddings are free under rate limits
768 dimensions are sufficient for your use case (most are)
You want to avoid single-provider dependency
You're building a new project and want to start cost-effectively

Try It: Migrate in 3 Lines

pip install schift

from schift import Schift
s = Schift(api_key="sk-...")

# Test migration quality before committing
report = s.bench.run(
  source="openai/text-embedding-3-small",
  target="google/gemini-embedding-2"
)
print(report)  # Recovery: 99.7% | Verdict: SAFE