OpenAI text-embedding-3-small API Pricing, Benchmarks & Rate Card (2026)
Complete 2026 developer pricing guide for OpenAI Text Embedding 3 Small. Benchmark token economics, context window limits, prompt caching discounts, and live inference latency across official provider endpoints.
⚡ 60-Second Budget Embedding Verdict✓ Verified October 2026
1536 Dimensions • 5x Cheaper than ada-002
OpenAI text-embedding-3-small Economics: OpenAI text-embedding-3-small provides rock-bottom vector embedding economics at just $0.02 per 1M tokens (5x cheaper than legacy text-embedding-ada-002). With 1536 dimensions and 8k context, it is the premier economic engine for high-scale document indexing, semantic caching, and initial candidate retrieval in multi-stage RAG pipelines. Cross-reference options in Cheapest Embedding APIs and compare with 3-Large Model.
Estimate your total bill based on prompt size, response volume, and request volume.
Estimated Monthly Bill
$0.00
Calculated using live 2026 rates for OpenAI Text Embedding 3 Small
Architectural Breakdown & Cost Optimization
OpenAI Text Embedding 3 Small represents a compelling value proposition in the 2026 developer ecosystem, balancing high accuracy with predictable operating margins. When designing agentic workflows or production RAG systems around OpenAI Text Embedding 3 Small, prompt engineering and caching strategy represent the single largest levers on your monthly bill.
To maximize unit economics:
Leverage Prefix Caching: Structure your system prompts and tool definitions at the beginning of the context payload to guarantee cache hits across multi-turn interactions.
Implement Batch Processing: For non-realtime operations such as data evaluation, document parsing, and backfill tasks, use asynchronous Batch APIs to immediately slash input and output bills in half.
Monitor Token Bloat: Uncontrolled conversational turns degrade both latency and token economics. Prune older history or summarize intermediate thoughts before context exceeds 32k tokens.
Frequently Asked Questions
What is the token pricing for OpenAI Text Embedding 3 Small in 2026?
OpenAI Text Embedding 3 Small is billed at $0.02 per million input tokens and $0.00 per million output tokens under standard API tiers.
What is the context window capacity of OpenAI Text Embedding 3 Small?
OpenAI Text Embedding 3 Small supports up to 8,191 tokens, allowing long-document ingestion and extensive conversational history.
Does OpenAI Text Embedding 3 Small support prompt caching or batch discounts?
Yes, prompt caching reduces recurring prefix token costs to roughly $0.00 per million tokens for cached blocks.
At $0.020 per million tokens, OpenAI's `text-embedding-3-small` is one of the most cost-effective dense embedding models available for high-throughput enterprise semantic search. Generating 1,536-dimensional vectors by default, it offers support for Matryoshka dimension shortening down to 512 dimensions. At 512 dimensions, vector memory consumption drops to 2.048 KB per document, allowing single-node Redis, PostgreSQL pgvector, or Milvus clusters to comfortably index over 5,000,000 document chunks within standard 16 GB RAM bounds.
Define an HNSW vector index optimized for 512-dimensional vectors using cosine distance:
CREATE TABLE document_chunks (
id BIGSERIAL PRIMARY KEY,
document_id UUID NOT NULL,
chunk_index INT NOT NULL,
content TEXT NOT NULL,
embedding vector(512)
);
CREATE INDEX idx_chunks_hnsw
ON document_chunks
USING hnsw (embedding vector_cosine_ops)
WITH (m = 16, ef_construction = 64);
High-Throughput Ingestion Concurrency Guardrails
OpenAI enforces strict Tokens Per Minute (TPM) limits across Tier 1 through Tier 5 developer accounts (from 1,000,000 TPM to 10,000,000 TPM). When running high-throughput ingestion pipelines, partition document arrays into batches of 500 chunks, and deploy an asynchronous token-bucket rate limiter with exponential jitter backoff to eliminate HTTP 429 exceptions during mass backfills.
When querying 512-dimensional text-embedding-3-small vectors in PostgreSQL using pgvector, setting `SET hnsw.ef_search = 40` provides an ideal balance of sub-8ms P99 search latency and 97% recall accuracy. In high-concurrency environments with over 1,000 queries per second, pre-filtering candidates using relational B-tree indexes before computing vector cosine distances cuts CPU utilization across PostgreSQL worker pools by over 60%.
For memory-constrained edge servers, storing vectors in INT8 scalar-quantized representations shrinks per-vector memory consumption to only 512 bytes, enabling multi-million document search on standard micro-instances with minimal recall degradation.
When scaling PostgreSQL pgvector databases beyond 10 million text-embedding-3-small vectors, table partitioning by creation date or tenant group prevents HNSW index bloat. Maintaining individual partition indexes below 2 GB guarantees that active index nodes fit entirely within shared_buffers RAM, preserving sub-10ms P95 query latencies.
For systems running automated nightly document syncs, running `VACUUM ANALYZE` on modified vector tables re-indexes candidate graphs and prevents recall drift over long operational lifecycles.
pgvector Connection Pooling & Read Scale-Out
When executing high-frequency cosine distance vector queries against PostgreSQL pgvector tables, routing read queries through PgBouncer connection poolers maintains sub-10ms query latencies and shields database backends from thread exhaustion during heavy user search spikes.