Mixedbread is a Berlin-based AI search and retrieval platform. It pairs open-weight embedding and reranking
models (mxbai-embed, Wholembed v3, mxbai-rerank v1/v2/v3-listwise) with a hosted REST API at
api.mixedbread.com that ships multimodal stores, document parsing, structured extraction, and connector-based
ingestion. Models are published on Hugging Face; the platform powers RAG and agent retrieval across text, PDFs,
images, audio, and video in 100+ languages.
URL: Visit APIs.json
Run: Capabilities Using Naftiko
- AI, Artificial Intelligence, Embeddings, Reranking, Search, Retrieval, RAG, Vector Database, Multimodal, Parsing, Open Weights
- Created: 2026-05-25
- Modified: 2026-05-25
| Model | Type | Modality | Parameters | Notes |
|---|---|---|---|---|
mixedbread-ai/wholembed-v3 |
Embedding | Omnimodal (text/image/audio/video), multilingual | — | Default store retriever; SOTA on LIMIT and BrowseComp-Plus (Mar 2026). |
mixedbread-ai/mxbai-embed-large-v1 |
Embedding | Text | 0.3B | Matryoshka + binary embeddings; 4.7M+ downloads on HF. |
mixedbread-ai/mxbai-embed-xsmall-v1 |
Embedding | Text | 24M | Edge-deployable. |
mixedbread-ai/deepset-mxbai-embed-de-large-v1 |
Embedding | German text | 0.5B | German-optimised, jointly with deepset. |
mixedbread-ai/mxbai-edge-colbert-v0-32m |
Late-interaction | Text | 32M | ColBERT-style for on-device retrieval. |
mixedbread-ai/mxbai-edge-colbert-v0-17m |
Late-interaction | Text | 17M | Smallest edge ColBERT variant. |
mxbai-rerank-v3-listwise |
Reranker | Text | — | Instruction-following listwise reranker (May 2026). |
mixedbread-ai/mxbai-rerank-large-v2 |
Reranker | Text | 2B | Strongest pointwise reranker. |
mixedbread-ai/mxbai-rerank-base-v2 |
Reranker | Text | 0.5B | Balanced size/quality. |
mixedbread-ai/mxbai-rerank-large-v1 |
Reranker | Text | — | v1 generation, broadly deployed. |
mixedbread-ai/mxbai-rerank-base-v1 |
Reranker | Text | 0.2B | — |
mixedbread-ai/mxbai-rerank-xsmall-v1 |
Reranker | Text | 71M | Lightweight inference. |
POST /v1/embeddings creates dense or multi-encoding embeddings for text and images. Supports Matryoshka
dimensions, binary/int8/float32 encodings, normalization, and prompt instructions.
Human URL: https://www.mixedbread.com/api-reference
POST /v1/reranking reorders candidate documents against a query using mxbai-rerank v1/v2 plus the
v3-listwise instruction-following reranker. Documents may be strings or structured objects with rank_fields.
Human URL: https://www.mixedbread.com/api-reference
Multimodal search indexes. CRUD on /v1/stores plus per-store file management, semantic search, regex grep,
chunk listing, question answering, query enhancement, metadata facets, search rules, and events. Replaces the
deprecated /v1/vector_stores surface (kept for backward compatibility).
Human URL: https://www.mixedbread.com/docs/stores/overview
- OpenAPI
- JSON Schema — Store
- Naftiko Capability — Stores CRUD
- Naftiko Capability — Store Files
- Naftiko Capability — Search & QA
- Example
Asynchronous document parsing as jobs. POST /v1/parsing/jobs starts a parse with configurable chunking
strategy and return format (markdown/HTML/text). Produces typed chunks for text, code, tables, images, audio,
and video.
Human URL: https://www.mixedbread.com/api-reference
Schema-guided structured extraction. POST /v1/extractions/jobs runs against a file; POST /v1/extractions/content
extracts from raw strings. /v1/extractions/schema create / enhance / validate JSON Schemas that drive the
extraction pipeline.
Human URL: https://www.mixedbread.com/api-reference
Workspace file storage shared across stores, parsing, and extraction. Single-shot POST /v1/files plus
multipart upload via /v1/files/uploads (create / complete / abort / list) and standard CRUD plus content
download.
Human URL: https://www.mixedbread.com/api-reference
POST /v1/chat/completions is an OpenAI-compatible chat surface that grounds responses in Mixedbread Stores.
Drives question-answering and agentic flows over indexed multimodal content.
Manage external data sources and their connectors, including OAuth2 authorization/callback flows. Used to ingest content from third-party SaaS systems into stores.
CRUD plus rotation (reroll) and revocation under /v1/api-keys.
Read-only billing telemetry — workspace summary, per-store costs, and parsing/store cost histograms. Maps to the FOCUS-aligned FinOps definition in this repo.
GET /v1/schemas/mxjson returns the canonical mxbai JSON schema for client validation.
Platform-administrator operations. POST /v1/admin/stores/{organization_id}/{store_identifier}/reingest
triggers a full tenant-store reingest.
- Python:
mixedbread-ai/mixedbread-python(current),mixedbread-ai/python-sdk(legacy) - TypeScript:
mixedbread-ai/mixedbread-ts(current),mixedbread-ai/typescript-sdk(legacy) - CLI: Mixedbread CLI for bulk document syncing —
mxbai stores sync - Agent Skills:
npx skills add mixedbread-ai/skillsfor Claude Code / Cursor / Claude Desktop
- LangChain —
mixedbread-ai-langchain - Haystack —
mixedbread-ai-haystack - LlamaIndex — Python and TypeScript forks under
mixedbread-ai/llama_indexandLlamaIndexTS - Vercel —
vercel-marketplace-mixedbread - Astro Starlight —
mixedbread-starlight - Next.js docs search —
nextjs-docs-searchandmxbai-docs-chat - Model Context Protocol — first-class MCP server for AI assistants
mxbai-rerank— Open reranker models and reference codebaguetter— Flexible sparse / dense / hybrid search librarybatched— Dynamic batching API for inference workloadsmaxsim-cpu— CPU late-interaction MaxSim kernelsmgrep— Semantic CLI grep (code, images, PDFs)astchunk— AST-aware code chunkingopenbread— Open-source collectionbinary-embeddings— 32x storage savings demo
| Plan | Price | Stores | Rate limit | Notes |
|---|---|---|---|---|
| Starter | Free + $5 one-time credit | 10 | 100 req/min | 3 workspace users, community Slack |
| Scale | $20/mo + $20 included credit | 10,000 | 1,200 query/min, 360 ingestion/min | Unlimited users, automatic backups, priority Slack |
| Enterprise | Custom | Unlimited | Custom | BYOC, dedicated support, PITR backups |
| Startup Program | Up to $250 one-time credit | — | — | VC-funded or raised up to $20M |
Usage-based pricing:
| Category | SKU | Unit | Price (USD) |
|---|---|---|---|
| Index — Fast | index_fast |
1M tokens | $1.50 |
| Index — High Quality | index_high_quality |
1M tokens | $3.00 |
| Search — Basic | search_basic |
1K queries | $4.00 |
| Search — Rerank | search_rerank |
1K queries | $7.50 |
| Search — Agentic | search_agentic |
1K queries | $20.00 |
| Storage | data_storage |
1M tokens / month | $0.50 |
See plans/mixedbread-ai-plans-pricing.yml,
rate-limits/mixedbread-ai-rate-limits.yml, and
finops/mixedbread-ai-finops.yml for machine-readable detail.
- Portal: https://www.mixedbread.com/
- Docs: https://www.mixedbread.com/docs
- API Reference: https://www.mixedbread.com/api-reference
- Pricing: https://www.mixedbread.com/pricing
- Blog: https://www.mixedbread.com/blog
- Platform: https://platform.mixedbread.com/
- GitHub: https://github.com/mixedbread-ai
- Hugging Face: https://huggingface.co/mixedbread-ai
- LinkedIn: https://www.linkedin.com/company/mixedbread-ai/
- X (Twitter): https://x.com/mixedbreadai
- Discord: https://discord.gg/mixedbread