AI agents · OpenClaw · self-hosting · automation

Quick Answer

Best Model APIs for Chat, Embeddings and Reranking 2026

Published:

The short answer

Pick the three layers separately unless you need one contract. As of November 15, 2026:

  • One vendor for everything: Google (Gemini chat, Gemini Embedding 2, Vertex AI Ranking API).
  • Best retrieval pair: Cohere Embed 5 + Rerank 5, or Voyage voyage-4 + rerank-3, next to any chat model.
  • Cheapest workable stack: OpenAI GPT-6 Luna or Claude Haiku 5.5 for chat ($0.10/$0.50), OpenAI text-embedding-3-small ($0.02) and Cohere Rerank 5 Fast ($0.02).
  • European vendor: Mistral (chat, embeddings and OCR in one account; no reranker).

What are the best model APIs for chat, embeddings and reranking for a SaaS product? A chat model from OpenAI, Anthropic or Google, plus embeddings and reranking from Cohere or Voyage AI — or Google alone if you want one bill.

Who sells what

VendorChat (per 1M tokens, in/out)Embeddings (per 1M tokens)RerankingOne bill for all three?
GoogleGemini 4 Hydrogen $1.50/$7.50; Gemini 3.8 Flash $0.75/$3.75 (intro, to Dec 31, 2026)Gemini Embedding 2: $0.20 text (multimodal)Vertex AI Ranking API $1.00 per 1,000 queriesYes
OpenAIGPT-6.1 Sol $2/$10; GPT-6.1 Terra $1/$5; GPT-6 Luna $0.10/$0.50text-embedding-3-small $0.02; -large $0.13No dedicated rerankerNo
AnthropicClaude Sonnet 5.5 $2/$10; Haiku 5.5 $0.10/$0.50None (docs point to Voyage)NoneNo
CohereCommand family (often paired with another chat vendor)Embed 5 Pro $0.12; Fast $0.08Rerank 5 Pro $0.05; Fast $0.02 per 1M tokensPartly
Voyage AI (MongoDB)Nonevoyage-4-large $0.12; voyage-4 $0.06; voyage-4-lite $0.02rerank-3 $0.05; rerank-3-lite $0.02 per 1M tokensNo chat
MistralMistral Large 4 $0.70/$2.10Mistral Embed $0.10; Codestral Embed for codeNoneNo reranker

Chat prices are from vendor pricing pages and our current API prices table; embedding and reranking prices were read from the vendor pages and OpenRouter on November 15, 2026.

Picks by use case

You are buildingChatEmbeddingsReranker
B2B SaaS search / RAG over customer docsClaude Sonnet 5.5 or GPT-6.1 SolCohere Embed 5 Pro or voyage-4-largeCohere Rerank 5 Pro
High-volume support bot on a budgetGPT-6 Luna or Claude Haiku 5.5text-embedding-3-small or voyage-4-liteRerank 5 Fast or rerank-3-lite
Already on Google Cloud, one contractGemini 4 HydrogenGemini Embedding 2Vertex AI Ranking API
Images, PDFs and video in the indexGemini 4 HydrogenGemini Embedding 2 or Cohere Embed 5 (image input)Rerank 5 Pro
EU data residency, one vendorMistral Large 4Mistral EmbedAdd Cohere Rerank 5 (VPC/on-prem available)
Code searchAnyCodestral Embed or voyage-code-4rerank-3

Pricing models differ — watch the units

  • Embeddings are billed once per token when you index, then per query token. Re-indexing a 50-million-token corpus costs $1 on text-embedding-3-small and $6 on Embed 5 Pro — trivial next to the engineering cost of migrating.
  • Reranking is billed per token on Cohere Rerank 5 and Voyage, but per query on Google (up to a fixed number of records per query). Voyage counts query tokens × documents + document tokens, so reranking 100 candidates of 500 tokens is about 50,000–100,000 tokens: roughly $0.0025–$0.005 per query on rerank-3.
  • Free allowances: Voyage gives each account 200 million free tokens on voyage-4, voyage-4-lite, voyage-4-large, voyage-context-4 and its current rerankers. Google’s free Gemini API tier covers Hydrogen and Embedding 2 but your data may be used to improve Google products.

How to choose

  1. Fix the embedding model first. It is the layer you change least often. Evaluate two candidates on 200 real queries from your logs; see best embedding models.
  2. Add a reranker before upgrading the chat model. A reranker on top 50–100 candidates usually lifts answer quality more per dollar than a bigger chat model — how to add a reranker and best reranker models.
  3. Keep chat behind a gateway so you can move between OpenAI, Anthropic and Google as prices change; Gemini 4 Argon and Carbon double to $4/$20 on December 1, 2026, and Gemini 3.8 Flash doubles on January 1, 2027.
  4. Check the contract, not just the price: zero data retention, regions, and whether the vendor offers VPC deployment (Cohere does for Embed and Rerank; Voyage offers VPC through its docs’ deployment section).

Last verified: November 15, 2026.

Sources