Billing
Pinecone is now available in Google Cloud Marketplace
You can now sign up for Pinecone billing through Google Cloud Marketplace.General
Organizations are generally available
Pinecone now features organizations, which allow one or more users to control billing and project settings across multiple projects owned by the same organization.Database
p2 pod type is generally available
The p2 pod type is now generally available and ready for production workloads. p2 pods are now available in the Starter plan and support the dotproduct distance metric.Database
Performance improvements
- Bulk vector deletes are now up to 10x faster in many circumstances.
- Creating collections is now faster.
DatabaseSearch
Hybrid search is in early access
Pinecone now supports keyword-aware semantic search with the new hybrid search indexes and endpoints. Hybrid search enables improved relevance for semantic search results by combining them with keyword search.This is an early access feature and is available only by signing up.General
Status page
The new Pinecone Status Page displays information about the status of the Pinecone service, including the status of individual cloud regions and a log of recent incidents.Database
Public collections
You can now create indexes from public collections, which are collections containing public data from real-world data sources. Currently, public collections include the Glue - SSTB collection, the TREC Question classification collection, and the SQuAD collection.Database
Collections are in public preview
You can now make static copies of your index using collections. After you create a collection from an index, you can create a new index from that collection. The new index can use any pod type and any number of pods. Collections only consume storage.This is a public preview feature and isn’t appropriate for production workloads.Database
Vertical scaling
You can now change the size of the pods for a live index to accommodate more vectors or queries without interrupting reads or writes. The p1 and s1 pod types are now available in 4 different sizes:1x, 2x, 4x, and 8x. Capacity and compute per pod double with each size increment.Database
p2 pod type is in public preview
The new p2 pod type provides search speeds of around 5ms and throughput of 200 queries per second per replica, or approximately 10x faster speeds and higher throughput than the p1 pod type, depending on your data and network conditions.This is a public preview feature and isn’t appropriate for production workloads.Database
API
Metadata filtering for describe_index_stats
You can now specify a metadata filter to get results for a subset of the vectors in your index by calling describe_index_stats with a filter object.The describe_index_stats operation now uses the POST HTTP request type. The filter parameter is only accepted by describe_index_stats calls using the POST request type. Calls to describe_index_stats using the GET request type are now deprecated.General
Pinecone console guided tour
You can now choose to follow a guided tour in the Pinecone console. This interactive tutorial walks you through creating your first index, upserting vectors, and querying your data. The purpose of the tour is to show you all the steps you need to start your first project in Pinecone.API
Updated response codes
Thecreate_index, delete_index, and scale_index operations now use more specific HTTP response codes that describe the type of operation that succeeded.Database
Selective metadata indexing
You can now store more metadata and more unique metadata values. Select which metadata fields you want to index for filtering and which fields you only wish to store and retrieve. When you index metadata fields, you can filter vector search queries using those fields. When you store metadata fields without indexing them, you keep memory utilization low, especially when you have many unique metadata values, and therefore can fit more vectors per pod.Database
Single-vector queries
You can now specify a single query vector using the vector input. We now encourage all users to query using a single vector rather than a batch of vectors, because batching queries can lead to long response messages and query times, and single queries execute just as fast on the server side.Database
Query by ID
You can now query your Pinecone index using only the ID for another vector. This is useful when you want to search for the nearest neighbors of a vector that is already stored in Pinecone.Database
Improved index fullness accuracy
The index fullness metric indescribe_index_stats results is now more accurate.Database
Partial updates are in public preview
You can now perform a partial update by ID and individual value pairs. This allows you to update individual metadata fields without having to upsert a matching vector or update all metadata fields at once.Database
New metrics
Users on all plans can now see metrics for the past one (1) week in the Pinecone console. Users on the Enterprise plan now have access to the following metrics via the Prometheus metrics endpoint:pinecone_vector_countpinecone_request_count_totalpinecone_request_error_count_totalpinecone_request_latency_secondspinecone_index_fullness(public preview)
pinecone_index_fullness metric is improved. This may result in changes from historic reported values. This metric is in public preview.SDK
Spark connector
Spark users who want to manage parallel upserts into Pinecone can now use the official Spark connector for Pinecone to upsert their data from a Spark dataframe.Database
Support for Boolean and float metadata in Pinecone indexes
You can now addBoolean and float64 values to metadata JSON objects associated with a Pinecone index.API
New state field in describe_index results
The describe_index operation results now contain a value for state, which describes the state of the index. The possible values for state are Initializing, ScalingUp, ScalingDown, Terminating, and Ready.