_This blog post was originally published on kero's [personal blog](https://blog.kerollmops.com/meilisearch-indexes-embeddings-7x-faster-with-binary-quantization)._

<Callout intent="note">
This is one blog post in a series:
- [Spotify-Inspired: Elevating Meilisearch with Hybrid Search and Rust](https://blog.kerollmops.com/spotify-inspired-elevating-meilisearch-with-hybrid-search-and-rust)
- [Multithreading and Memory-Mapping: Refining ANN Performance with Arroy](https://blog.kerollmops.com/multithreading-and-memory-mapping-refining-ann-performance-with-arroy-in-rust)
- [Meilisearch Expands Search Power with Arroy's Filtered Disk ANN](https://blog.kerollmops.com/meilisearch-expands-search-power-with-arroy-s-filtered-disk-ann)
- [How Meilisearch Updates a Millions Vector Embeddings Database in Under a Minute](https://blog.kerollmops.com/how-meilisearch-updates-a-millions-vector-embeddings-database-in-under-a-minute)
- Meilisearch Indexes Embeddings 7x Faster with Binary Quantization.
</Callout>

For the past year at Meilisearch, [we’ve been working hard on hybrid search](https://github.com/meilisearch/meilisearch/issues/3838), which mixes keyword search with semantic search using [Arroy: our vector store, based on a Spotify technology](https://blog.kerollmops.com/spotify-inspired-elevating-meilisearch-with-hybrid-search-and-rust). A vector store is a data structure that efficiently stores embeddings (vectors) for fast and relevant retrieval, called Approximate Nearest Neighbors (ANN) search.

When our customers started to use arroy heavily, some of them started reaching the machine limits. For example, for a customer with a 768 dimensions model, we found out that we couldn’t index more than 15M of embeddings on a machine with 64GiB of RAM.

### Binary Quantization to the Rescue

The concept behind this elaborate term is that we are going to *quantize* the embeddings. Quantizing a number involves spreading it into a defined number of values; for *binary* quantization, two values can be represented on a single bit, e.g., 0.2561 becomes 1, and -0.568 becomes -1.

This implies that we could convert a 32-bit float number into a 1-bit number, which would divide the disk and RAM usage by 32. That would mean we could index up to 480M embeddings with 64GiB of RAM, instead of being limited to 15M embeddings.

An issue we faced is that bits cannot represent -1 and 1 but only 0 and 1. We had to use some tricks to make it work and fake 0s into -1s in our computations. For example, one of the functions that we had to optimize a lot was the function in charge of converting a binary quantized vector to 32-bit float vectors.

We do not publicly expose raw quantized embeddings and utilize this function to provide real 32-bit float embeddings to the users. However, the primary reason for converting 1-bit quantized embeddings to 32-bit floats is to ensure high precision when [generating hyperplanes to group embeddings together](https://blog.kerollmops.com/spotify-inspired-elevating-meilisearch-with-hybrid-search-and-rust#spotify-s-hyperplane-trees-for-efficient-anns) and maintaining relevant ANNs.

![Profiling backtrace showing slow 1-bit to 32-bit float conversion](https://unable-actionable-car.media.strapiapp.com/68747470733a2f2f692e696d6775722e636f6d2f497165713649562e706e67_596adc5fd5.png)

_Converting a quantized embedding back to 32-bit floating points takes up to 24.21s on a 1m53s run._

Below is the SIMD neon-specific version of this function. It processes raw bytes corresponding to a 1-bit quantized embedding, converting them into a properly aligned 32-bit float embedding. Using masks, the function separates the low and high bits, processing them in batches of four in 128-bit registers, which are then transformed into four 32-bit floats. These values are collectively written to the output vector.

```rust
unsafe fn to_vec_neon(bytes: &[u8]) -> Vec<f32> {
    let mut output: Vec<f32> = vec![0.0; bytes.len() / u32::BITS];
    let output_ptr = output.as_mut_ptr();
    let low_mask = [0b_0000_0001, 0b_0000_0010, 0b_0000_0100, 0b_0000_1000];
    let high_mask = [0b_0001_0000, 0b_0010_0000, 0b_0100_0000, 0b_1000_0000];
    let ones = unsafe { vld1q_dup_f32(&1.0) };
    let minus = unsafe { vld1q_dup_f32(&-1.0) };

    for (current_byte, base) in bytes.iter().enumerate() {
        unsafe {
            let base = *base as u32;
            let base = vld1q_dup_u32(&base);
            for (i, mask) in [low_mask, high_mask].iter().enumerate() {
                let mask = vld1q_u32(mask.as_ptr());
                let mask = vandq_u32(base, mask);
                // 0xffffffff if equal to zero and 0x00000000 otherwise
                let mask = vceqzq_u32(mask);
                let lane = vbslq_f32(mask, minus, ones);
                let offset = output_ptr.add(current_byte * 8 + i * 4);
                vst1q_f32(offset, lane);
            }
        }
    }

    output
}
```

_<a href="https://x.com/lemire/status/1835700368648323261">Daniel Lemire doesn't seem to have found a better algorithm</a>_ 🤭

### The Results

We chatted about the theoretical benefits of this approach earlier, but how does it work out in real life? Here are different tables summarizing the results for you.

<details>
  <summary>Open to see the original benchmarks</summary>
  <img src="https://unable-actionable-car.media.strapiapp.com/68747470733a2f2f692e696d6775722e636f6d2f67667765626b592e706e67_a96a995c7c.png"  alt="Original arroy benchmark results"/>
</details>

#### Embeddings with 1024 Dimensions

When examining small embeddings, we can observe a significant impact on relevancy due to binary quantization. The primary advantages of this approach lie in faster insertion times and reduced storage requirements. It is not advisable to enable this feature for small embeddings, as they already occupy minimal disk space. Storing over four million 1024-dimensional embeddings requires just 8 GiB of disk space.

| Version                 | Recall @1 | Recall @20 | Recall @100 | Indexing time | Search time | On disk size |
|-------------------------|-----------|------------|-------------|---------------|-------------|--------------|
| Cosine                  | **1.00**  | **0.72**   | **0.84**    | 491s          | **23s**     | 8 GiB        |
| Binary Quantized Cosine | 0.87      | 0.50       | 0.52        | **147s**      | 27s         | **850 MiB**  |

#### Embeddings with 1536 Dimensions

Binary quantization proves advantageous for managing larger embeddings. It significantly reduces search times, utilizing just 15% of the original disk space. Indexing about 500,000 1536-dimensional embeddings is achieved in a quarter of the time. Despite this, there remains a noticeable impact on relevancy...

| Version                 | Recall @1 | Recall @20 | Recall @100 | Indexing time | Search time | On disk size |
|-------------------------|-----------|------------|-------------|---------------|-------------|--------------|
| Cosine                  | **1.00**  | **0.82**   | **0.90**    | 1219s         | 50s         | 13 GiB       |
| Binary Quantized Cosine | **1.00**  | 0.70       | 0.70        | **235s**      | **47s**     | **2 GiB**    |

#### Embeddings with 3072 Dimensions

The impact of binary quantization on large embeddings is evident. While relevance may be slightly affected, the reduction in disk usage, search time, and indexing time is significant. Being able to store the same number of embeddings with just 15% of the space and index 6 times faster is a major achievement! It takes half as much disk space to store a million 3072-dimensional binary quantized embeddings than four million 1024-dimensional embeddings.

| Version                 | Recall @1 | Recall @20 | Recall @100 | Indexing time | Search time | On disk size |
|-------------------------|-----------|------------|-------------|---------------|-------------|--------------|
| Cosine                  | **1.00**  | **0.86**   | **0.92**    | 4000s         | 150s        | 26 GiB       |
| Binary Quantized Cosine | **1.00**  | 0.77       | 0.77        | **560s**      | **100s**    | **4 GiB**    |

### Conclusion

The impact of binary quantization on recall was expected to decrease, given the reduced information in the algorithm. However, for datasets with over 1500 dimensions, the impact on recall is relatively minor compared to the significant benefits in disk space and indexing speed.

In practice, we observed that space utilization decreased not by 32 times but by just 15% of the original space, a notable achievement. This is primarily due to arroy storing not only user vectors but also internal nodes for efficient searching. Split nodes (hyperplanes) are retained as full 32-bit floats to maintain precision.

- Indexing speed improves nearly 5 times with 1536 dimensions and about 7 times with 3072 dimensions.
- Disk size shrinks to around 6.5 times smaller.
- Search time shows slight improvements for embeddings starting at 1536 dimensions.

We recommend using Meilisearch/arroy's binary quantization feature for OpenAI users with large embeddings. In an upcoming article, we will compare Meilisearch/arroy with competitors, focusing on our Filtered DiskANN feature. Stay tuned!

Meilisearch indexes embeddings 7x faster with binary quantization

We've been busy working on things you've asked for, and the team is excited to share what's new in Meilisearch this quarter.

## A revamped search preview 🔍

The [search preview](https://meilisearch.com.cn/docs/learn/getting_started/search_preview) has been one of our most-used features, but it didn't let you test everything Meilisearch can do. We've now rebuilt our search preview to make it more useful to assess test search result relevancy when you are tweaking an index's settings.

What's included:

- Show/hide the results' [ranking score](https://meilisearch.com.cn/docs/learn/relevancy/ranking_score):

![ranking score.png](https://unable-actionable-car.media.strapiapp.com/image_38_9583f10fe8.png)

- Check and customize search options
- Configure filterable attributes as facets
- Test [AI-powered searches](https://meilisearch.com.cn/docs/learn/ai_powered_search/getting_started_with_ai_search) if this functionality has been enabled for your project.

An updated preview allows you to test everything in real-time, making it easier to get your search working exactly how you want.

## New index settings in the UI ⚙️

We've fixed one of your biggest frustrations - you and your team no longer need to rely on JSON input to [change your index settings](https://meilisearch.com.cn/docs/learn/configuration/configuring_index_settings). Instead of typing JSON code, you'll now see straightforward controls in the UI for adjusting settings like:

- Add and delete searchable attributes:
![searchable attributes/index settings.png](https://unable-actionable-car.media.strapiapp.com/image_37_5bea152778.png)

- Change the ranks of your searchable attributes
- Reset to default

Now anyone on your team can get access to user-friendly modification of the index settings without needing to understand JSON syntax.

## Speed up your search  with feature opt-out 🚀

We've got some updates for those of you working with massive datasets. To **reduce disk usage** and **speed up indexing by up to 15%**, you can now optimize Meilisearch's performance: 

- Choose to opt-out of prefix search and facet search features
- Reduce indexing time and save disk space

This has been one of our most requested features from users dealing with large-scale deployments, and we're excited to see how you'll use it.

## What else is cooking? 🎯
#### New Japan region is coming in Q4! 🇯🇵
Soon, you’ll be able to host your Meilisearch projects in Japan. This means much faster searches for users in Asia-Pacific - instead of connecting to Singapore, they'll connect directly to servers in Japan. The difference? A search that used to take 80+ milliseconds will happen almost instantly.

To use it, you’ll just need to pick Japan when you create a new project in the [cloud](https://cloud.meilisearch.com) dashboard. Keep an eye on our updates to know exactly when it’s released!


We are not stopping here. Our team is already working on our early 2025 releases. Here's a quick spoiler:

- AI-powered hybrid search general availability
- New Cloud dashboard features
- Even better performance optimizations

## Try it out!

Want to test these features? [Sign up for Meilisearch Cloud](https://cloud.meilisearch.com/) to experience these new features firsthand.

---

We built these changes based on your feedback, and we'd love to know if they help. Try them out and tell us what you think.

For  a more in-depth look at these updates, visit our [documentation](https://docs.meilisearch.com/).

Meilisearch Q4 2024 release wrap-up

##  🐿️ Meilisearch 1.11 release

![Release notes - Meilisearch 1.11](https://unable-actionable-car.media.strapiapp.com/meilisearch_1_11_59ea013828.jpg)

Meilisearch 1.11 introduces key **improvements** in **AI-powered search**, including **binary quantization**. This release also **enhances federated search** functionality based on user feedback and introduces a **new filter operator**.

[Discover updates](https://meilisearch.com.cn/blog/meilisearch-1-11)

***


## 🌎 TutKit's journey with Meilisearch 

[![Meilisearch powers multilingual learning at scale for Tutkit](https://unable-actionable-car.media.strapiapp.com/Tut_Kit_516315e429.png)](https://meilisearch.com.cn/blog/tutkit-multilingual-learning-at-scale)

Discover how Tutkit.com scaled its learning platform to handle search across **15,000+ resources in 26 languages**.

[Read case study](https://meilisearch.com.cn/blog/tutkit-multilingual-learning-at-scale)

***

## 🎥 Webinar recording
[![October 9th.png](https://unable-actionable-car.media.strapiapp.com/October_9th_b09a084f13.png)](https://www.youtube.com/watch?v=CvutT64u0fE)


Missed our October event? No worries—we've got you covered with the full recording!

 

Discover Meilisearch's **roadmap** and latest **updates**, learn more about **federated search and its benefits** and see a live demo of the **new Cloud functionality**.

[Watch it now!](https://www.youtube.com/watch?v=CvutT64u0fE)

***

## 📰 In the news
### ⚛️  [Add AI-powered search to React](https://meilisearch.com.cn/blog/add-ai-powered-search-to-react)
Build a React movie search and recommendation app with Meilisearch's AI-powered search.

### 🔍 [Federated search guide](https://meilisearch.com.cn/docs/learn/multi_search/performing_federated_search)

Learn how to create separate indexes for different CRM data types and query them simultaneously for a unified results list.

### ✨ [Semantic search with Cloudflare embeddings](https://meilisearch.com.cn/docs/guides/ai/cloudflare)

Learn how to set up Meilisearch with Cloudflare Worker AI embeddings to enable semantic search capabilities.

### 🛍️ [Shopify storefront guide for Next.js](https://meilisearch.com.cn/blog/next-js-shopify-storefront-with-blazity)

Learn to build a Shopify storefront with Next.js and the [Blazity commerce starter](https://github.com/Blazity/enterprise-commerce), using Meilisearch hybrid search to serve all front-end data.

### 🔒 [Laravel multitenancy guide](https://meilisearch.com.cn/docs/guides/security/laravel_multitenancy)

How to implement search in a multitenant Laravel application.

***

## 🤗 Made with Meilisearch

A community member built [Firebits.io](https://firebits.io) and shared it with the Discord community. Firebits is a modern online Second Brain that helps you organize and share your online learning resources. Its search is, of course, powered by Meilisearch 🎉 

![firebits-2.gif](https://unable-actionable-car.media.strapiapp.com/firebits_2_869f9caaa0.gif)

We’re always looking to promote new projects from the community. Post yours in the [#built-with-meilisearch](https://discord.meilisearch.com/) channel!

***

## 🦸‍♀️ We’re hiring!

See our open positions:

- [Chief of Staff](https://jobs.lever.co/meili/4abb7650-a5ca-4aac-9f35-e2fe5144b78d)

***

## 🤓 Did you know?

The fastest way to try Meilisearch is our [Cloud free trial](https://meilisearch.com.cn/cloud).

[Get started!](https://meilisearch.com.cn/cloud)

***

Stay in the loop of everything Meilisearch by subscribing to our [monthly newsletter](https://dub.sh/meili-newsletter#newsletter&utm_source=blog&utm_medium=october-24-updates). To learn more about Meilisearch's future and help shape it, take a look at our [roadmap](https://roadmap.meilisearch.com/) and participate in our [Product Discussions](https://github.com/orgs/meilisearch/discussions).

For anything else, join our developers community on [Discord](https://discord.meilisearch.com/).



Meilisearch October updates

[TutKit](https://www.tutkit.com/en) is a comprehensive multilingual learning and content platform that offers training materials, marketing resources, and creative assets to its subscribers worldwide. Built by German agency [4eck Media](https://4eck-media.de/), the platform stands as a testament to the power of effective search functionality in educational platforms. 

**TutKit self-hosts Meilisearch to power high-performance search functionality across their Laravel-based learning platform, managing 15,000+ multilingual content pieces with millisecond response times.**

![TutKit.gif](https://unable-actionable-car.media.strapiapp.com/Tut_Kit_f4b038e4a6.gif)


>"With Meilisearch, we have chosen the right company for the perfect search." - Matthias Petri, founder of TutKit.com and managing director of [4eck Media](https://4eck-media.de/).

The challenge: scaling search for multilingual content
---------

TutKit.com's initial search functionality, implemented by an external agency using MySQL, suffered from severe performance issues due to inefficient programming. With the platform hosting tens of thousands of content pieces, users endured unacceptable loading times of up to 10 seconds for search results—a critical flaw for a content platform. The system also lacked proper multilingual support, making it impossible to effectively filter and display results across different languages.

Frustrated with these limitations, the TutKit team decided to take matters into their own hands. They began a comprehensive refactoring project, bringing the development in-house with a clear goal: to create an ultra-fast search system that could handle multiple languages.


The solution: implementing Meilisearch
---------
The search for a solution that would provide complete control over search results led Tutkit to Meilisearch. While the platform's modern UX and compelling website initially caught their attention, the final decision was driven by several additional factors:
- **Lightning-fast search performance**: while the platform's modern UX and compelling website initially caught their attention, it was Meilisearch's exceptional speed that stood out—presenting a dramatic improvement over their existing system
- **Robust multilingual support.** Meilisearch powers a fast search performance across 26  languages on TutKit platform, coupled with straightforward management capabilities.
- **Regular platform updates and improvements.** Meilisearch’s regular monthly updates continue to bring valuable improvements, enhancing the experience for both the internal TutKit team and the website visitors.

Technical implementation
---------
The TutKit development team executed a comprehensive implementation strategy:

 **1. System setup:** the implementation of Meilisearch began with a straightforward server installation from the official website. To ensure reliable operation, the team configured a system to manage the Meilisearch instance, including automatic startup and failure recovery. This configuration guarantees continuous service availability, even after server restarts.
 
**2. Security:** the instance was implemented through a dual-key system: 
- a master key for comprehensive backend access, which allows full access to the Meilisearch instance (including creation and index management), is used exclusively in the backend. 
- A custom API key with limited permissions for front-end operations. 
This separation creates a robust security layer, protecting backend operations while maintaining efficient front-end interactions. The Laravel application connects to Meilisearch using the API key, effectively minimizing the risk of unauthorized access and potential data breaches.

**3. Optimized data flow:** built on the Meilisearch PHP library, TutKit's custom Laravel integration ensures only necessary changes reach the search index. This selective approach minimizes data processing overhead while maintaining complete control over the implementation, resulting in streamlined index updates and improved search performance.

**4. Smart batch processing:** A custom Laravel console command handles large-scale updates and initial indexing processes. By targeting only relevant records and executing selectively for major changes, this tool significantly reduces server load and maintains system reliability through efficient resource utilization.

**5. Real-time synchronization:** The system automatically detects and processes changes in Laravel models, updating Meilisearch indices instantly without full reindexing. This targeted approach keeps search data synchronized with minimal resource usage, ensuring fast, responsive search functionality for users.

Impact and results
---------
The implementation of Meilisearch has delivered significant benefits for TutKit:
**Performance improvements:** thanks to Meilisearch's continuously improving indexing speed, TutKit can process new data into their search index with remarkable efficiency. The platform's lightning-fast search capabilities significantly reduce server load by eliminating the need for resource-intensive MySQL queries. 

**Real-time filtering capabilities** further enhance the system's responsiveness, allowing users to instantly refine and view their search results.

**Cost benefits:** adoption of Meilisearch as an open-source solution has delivered cost advantages for TutKit. Beyond eliminating ongoing licensing fees typically associated with paid providers, the platform's intuitive nature allowed TutKit's in-house development team to handle implementation and maintenance independently, avoiding external developer costs.

**User experience impact:** improved search functionality has directly enhanced the visitor experience on TutKit's platform. Fast, accurate search results have eliminated user frustration caused by long wait times and irrelevant results. This enhanced user experience has proven valuable for business outcomes, as satisfied users are more likely to convert into paying customers.

**Scalability success:** despite managing hundreds of thousands of content pieces across multiple languages, the platform maintains its exceptional performance. Meilisearch consistently delivers precise, relevant results while handling TutKit's growing content library and language variants. This robust scalability gives the team confidence in their ability to expand the platform's capabilities over the long term.

Looking forward
---------

TutKit continues to scale its platform, adding more content and language variants with confidence in Meilisearch's ability to handle growing demands. The team looks forward to the further development of [AI-powered features](https://meilisearch.com.cn/solutions/hybrid-search) that could further enhance search intelligence and simplify the user interface by reducing the need for extensive filtering options.

Matthias Petri, founder of TutKit.com and managing director of [4eck Media](https://4eck-media.de/) said: "A good search is like a stage for large content platforms, where exactly what the user wants to see is always played. Without Meilisearch, our platform would be just one of many - with mediocre search results and without the clear added value that we would like to offer our customers."

For TutKit, Meilisearch has proven to be more than just a search solution – it's become a cornerstone of their platform's success, enabling them to deliver on their promise of effective learning and productive application for users across the globe.


TutKit's journey with Meilisearch: powering multilingual learning at scale.

We're excited to introduce Meilisearch v1.11. In this article, we'll cover the most significant changes. For a full list of updates, visit the [changelog](https://github.com/meilisearch/meilisearch/releases/tag/v1.11.0) on GitHub.

<Callout>
  Meilisearch 1.11 is available on [Meilisearch Cloud](https://cloud.meilisearch.com/#release-v1-11&utm_source=blog), too—upgrade now!
</Callout>

## Experimental: vector database indexing performance improvements

Meilisearch v1.11 introduces a [new embedder option](https://meilisearch.com.cn/docs/reference/api/settings#release-v1-11&utm_source=blog#binaryquantized): `binaryQuantized`.

Binary quantization compresses vectors by encoding each dimension with 1-bit values, significantly boosting performance, especially in datasets with over 1M documents and models with more than 1024 dimensions.

Smaller embeddings (under 1024 dimensions) tend to give worse results. With fewer dimensions, the binary vector doesn't retain enough information for good relevancy.


<Callout>
Enabling this feature permanently updates all vectors to contain only 1s and -1s, which can greatly affect search relevancy. This change is irreversible.
</Callout>

Before enabling this feature, we recommend testing it in a smaller or duplicate index in a development environment to evaluate its impact on your use case.

## Experimental: AI-powered search improvements

### Breaking changes

This release introduces several breaking changes for feature stabilization. 

####  `embedder` parameter becomes mandatory

In previous versions, the default embedder was automatically used for search and similar requests when no embedder was specified. Starting with v1.11, it’s now mandatory to specify the `embedder` parameter when performing:
- Semantic or hybrid search via [`GET`](https://meilisearch.com.cn/docs/reference/api/search#release-v1-11&utm_source=blog#search-in-an-index-with-get) and [`POST`](https://meilisearch.com.cn/docs/reference/api/search#release-v1-11&utm_source=blog#search-in-an-index-with-post) where the embedder must be included as part of the `hybrid` object.
- Similarity search via [`GET`](https://meilisearch.com.cn/docs/reference/api/similar#release-v1-11&utm_source=blog#get-similar-documents-with-get) and [`POST`](https://meilisearch.com.cn/docs/reference/api/similar#release-v1-11&utm_source=blog#get-similar-documents-with-post), where the `embedder` is required as a top-level parameter.


```javascript
// Semantic and hybrid search
client.index('INDEX_NAME').search(
  query, {
    hybrid: {
      embedder: 'EMBEDDER_NAME' // mandatory as of v1.11
   }
  }
)

// Similarity search
client.index('INDEX_NAME').searchSimilarDocuments({ 
  id: 'TARGET_DOCUMENT_ID',
  embedder: 'EMBEDDER_NAME' // mandatory as of v1.11
})
```

Additionally, even for full-vector searches (when using `vector` without `q`), you must include the `hybrid` object with the `embedder`.

```javascript
client.index('INDEX_NAME').search({
  vector: [0, 1, 2],
  hybrid: {
      embedder: 'EMBEDDER_NAME'// mandatory as of v1.11
   }
})
```

#### Non-zero `semanticRatio` now ignored when passing `vector` without `q`

Before Meilisearch v1.11, when performing a search with the [`vector`](https://meilisearch.com.cn/docs/reference/api/search#release-v1-11&utm_source=blog#vector-experimental) and [`hybrid`](https://meilisearch.com.cn/docs/reference/api/search#release-v1-11&utm_source=blog#hybrid-search-experimental) parameters but without [`q`](https://meilisearch.com.cn/docs/reference/api/search#release-v1-11&utm_source=blog#query-q), a [placeholder keyword search](https://meilisearch.com.cn/docs/reference/api/search#release-v1-11&utm_source=blog#placeholder-search) occurred unless `hybrid.semanticRatio` was set to `1.0`. In v1.11, a semantic search is always performed unless `hybrid.semanticRatio` is set to `0.0`.

### Additions and improvements

#### Default OpenAI model update

The default [OpenAI model](https://meilisearch.com.cn/docs/reference/api/settings#release-v1-11&utm_source=blog#model) has been updated from `text-embedding-ada-002` to the new `text-embedding-3-small`. This new model is both more cost-effective and more accurate than the legacy `ada` model, offering better performance at a lower cost.

#### New embedder parameter

Meilisearch 1.11's new `documentTemplateMaxBytes` [parameter](https://meilisearch.com.cn/docs/reference/api/settings#release-v1-11&utm_source=blog#documenttemplatemaxbytes) allows truncating the document's template text when it exceeds the specified byte limit. It is set, by default, to 400 bytes. 

This change is backward compatible. Existing embedders will retain the 400-byte default, with no need to recompute embeddings during the upgrade. The new limit will only apply when a document is modified and the template is re-rendered.

#### Improved document template

Fields in `documentTemplate` now include the `field.is_searchable` property. The default [document template](https://meilisearch.com.cn/docs/reference/api/settings#release-v1-11&utm_source=blog#documenttemplate) will filter out both empty fields and fields not in the searchable attributes list. The new default template only applies to embedders created starting with v1.11.


```liquid
// v1.11 default document template
{% for field in fields %}
  {% if field.is_searchable and not field.value == nil %}
    {{ field.name }}: {{ field.value }}
  {% endif %}
{% endfor %}

// v1.10 default document template
{% for field in fields %}
  {{ field.name }}: {{ field.value }}
{% endfor %}
```

## Federated search improvements: facet distribution and stats

Meilisearch v1.11 introduces two new options for federated searches: `facetsByIndex` and `mergeFacets`. These options allow you to request facet [distribution](https://meilisearch.com.cn/docs/reference/api/search#release-v1-11&utm_source=blog#facetdistribution) and [stats](https://meilisearch.com.cn/docs/reference/api/search#release-v1-11&utm_source=blog#facetstats) for federated search results.

### Facet Information by index

To retrieve facet distribution and stats for each individual index, use the `facetsByIndex` [option](https://meilisearch.com.cn/docs/reference/api/multi_search#release-v1-11&utm_source=blog#facetsbyindex). 


```json
{
  "federation": {
    "facetsByIndex": {
      "movies": ["title", "id"],
      "comics": ["title"]
    }
  },
  "queries": [
    { "q": "Batman", "indexUid": "movies" },
    { "q": "Batman", "indexUid": "comics" }
  ]
}
```

The response will include facet distribution and stats for each index:


```json
{
  "facetsByIndex": {
    "movies": {
      // Facet distribution for queries on the `movies` index
      "distribution": {
        "title": {
          "Batman returns": 1
        },
        "id": {
          "42": 1
        }
      },
      // Facet stats for queries on the `movies` index
      "stats": {
        "id": {
          "min": 42,
          "max": 42
        }
      }
    },
    "comics": {
      // Facet distribution for queries on the `comics` index
      "distribution": {
        "title": {
          "Batman: the killing joke": 1
        }
      },
      // Facet stats for queries on the `comics` index (empty because only numeric values are considered)
      "stats": {}
    }
  }
}
```

### Merged facet information

To get facet distribution and stats from all indexes in a single result, use both [`facetsByIndex`](https://meilisearch.com.cn/docs/reference/api/multi_search#release-v1-11&utm_source=blog#facetsbyindex) and [`mergeFacets`](https://meilisearch.com.cn/docs/reference/api/multi_search#release-v1-11&utm_source=blog#mergefacets):


```json
POST /multi-search
{
  "federation": {
    "facetsByIndex": {
      "movies": ["title", "id"],
      "comics": ["title"]
    },
    "mergeFacets": { "maxValuesPerFacet": 10 }
  },
  "queries": [
    { "q": "Batman", "indexUid": "movies" },
    { "q": "Batman", "indexUid": "comics" }
  ]
}
```

The response will include merged facet data:


```json
{
  "facetDistribution": {
    "title": {
      "Batman returns": 1,
      "Batman: the killing joke": 1
    },
    "id": {
      "42": 1
    }
  },
  "facetStats": {
    "id": {
      "min": 42,
      "max": 42
    }
  }
}
```

<Callout>
Learn more about [federated search and its use cases](https://meilisearch.com.cn/blog/what-is-federated-search#release-v1-11&utm_source=blog).
</Callout>

## Experimental: new `STARTS_WITH` filter operator

The new `STARTS_WITH` [filter operator](https://meilisearch.com.cn/docs/learn/filtering_and_sorting/filter_expression_reference#release-v1-11&utm_source=blog#starts-with) matches any field values that start with the specified string pattern:


```json
// returns all heroes whose name starts with 'super'
"filter": "hero STARTS_WITH super"
```

This is an experimental feature. To enable it, use the [experimental features](https://meilisearch.com.cn/docs/reference/api/experimental_features#release-v1-11&utm_source=blog) endpoint and set the `containsFilter` option to `true`. Note that this will also activate the `CONTAINS` filter operator.

<Callout>
Share your feedback via this [GitHub discussion](https://github.com/orgs/meilisearch/discussions/763)
</Callout>

## Contributors shout-out

We want to give a massive thank you to the contributors who made this release possible: [@iornstein](https://github.com/iornstein) for their help with [Meilisearch](https://github.com/meilisearch/meilisearch), and [@luflow](https://github.com/luflow), [@tkhshtsh0917](https://github.com/tkhshtsh0917) for their work on [charabia](https://github.com/meilisearch/charabia).

We also want to extend special thanks to our SDK maintainers working hard to make Meilisearch available across many [languages and frameworks](https://meilisearch.com.cn/docs/learn/what_is_meilisearch/sdks#release-v1-11&utm_source=blog).

* * *

And that’s a wrap for v1.11! These release notes only highlights the most significant updates. For an exhaustive listing, read the [changelog](https://github.com/meilisearch/meilisearch/releases/tag/v1.11.0) on Github.

Upgrade with ease using Meilisearch Cloud.

[Open Meilisearch Cloud](https://cloud.meilisearch.com/login#release-v1-11&utm_source=blog)

* * *

Stay in the loop of everything Meilisearch by subscribing to our [monthly newsletter](https://dub.sh/meili-newsletter#release-v1-11&utm_source=blog). To learn more about Meilisearch's future and help shape it, take a look at our [roadmap](https://roadmap.meilisearch.com/#release-v1-11&utm_source=blog) and participate in our [Product Discussions](https://github.com/orgs/meilisearch/discussions).

For anything else, join our developers community on [Discord](https://discord.meilisearch.com/#release-v1-11&utm_source=blog).

Meilisearch 1.11

_Bildhistoria chose Meilisearch to power the search through its quickly growing database of historical photographs and affiliated data, driving the search from multiple sources of metadata and mapsearch._

> _“Bildhistoria's large volumes of historical photos, persons, places and organizations require a fast and easy to maintain search engine, like Meilisearch. Meilisearch is used on so many places around Bildhistoria, enabling our users to explore photos and learn more about ancestors and local history. This makes Meilisearch instrumental in the success of our community.”_ - Per Rohdin, board member of the community, and project manager at Bildhistoria.

[**Bildhistoria**](https://bildhistoria.se/)**, a Swedish non-profit, operates the country's largest collection of copyright-free historical photographs. It focuses on collecting and preserving historical photographs from museums and private collections. While the project has not yet reached full general availability, Bildhistoria already boasts an impressive archive of around 90,000 images dating from the 1950s and earlier.**

The project aims to create a dynamic and collaborative platform where the public can interact with history, and where professional and amateur genealogists alike can contribute by enriching the photographs with additional data.

**Challenge: Bildhistoria tackles data fragmentation and community engagement**
-------------------------------------------------------------------------------

From the start, Bildhistoria faced three main challenges in bringing their vision to life:

**Challenge: searching across multiple types of data:** Bildhistoria's main challenge with search is effectively searching across the different data types. The platform features not only photographs but also information about people’s names, organizations, and locations associated with each image. The team needed a search solution that could seamlessly navigate through these different data types. To address this, they have implemented Meilisearch’s "[multi-index search](https://meilisearch.com.cn/docs/reference/api/multi_search)" that returns results from all the different indexes, allowing users to find relevant information regardless of whether they are searching for a person, place, or organization.

**Challenge: data scattered across multiple sources:** Historical photographs were scattered across various sources, including museums, archives, and private collections. There was no centralized platform in Sweden to consolidate these copyright-free images, making it difficult for researchers and the public to access this wealth of visual history.

![](https://unable-actionable-car.media.strapiapp.com/AD_4n_Xf_R_Ormjkncn9_Uj8i_Un6vaqs_Wi_I_Uln1hutwuz0h_KI_wm_GIA_P_Wry5j_Rj5_R_Altplg_VV_5_Jy_ANKPL_2_Hayt3_Jv_Ku_A_Gw9ou_P8l0va_9f26862acc.jpg)

**Challenge: Community accessibility and engagement:** Unlike traditional digital archives, Bildhistoria is an interactive platform where users can contribute additional historical facts and enrich the metadata associated with the photographs. The community of genealogists is large and diverse, spanning professional organizations to amateur genealogists and the general public. Bildhistoria’s team knew from the beginning that the only way to make the project work was to make it easily accessible and easy to use for everyone.

**Solution: Bildhistoria chose Meilisearch for simplified operations, team-wide accessibility, and rapid data synchronization**
-------------------------------------------------------------------------------------------------------------------------------

After careful consideration, Bildhistoria chose Meilisearch as their search solution for several key reasons:

**Less operational overhead:** Meilisearch proved much easier to set up and connect to existing indexes compared to alternatives like Elasticsearch. This was crucial for a non-profit organization with limited resources.

**Accessibility for all team members:** Meilisearch's intuitive admin interface allowed even non-technical staff to understand and configure search settings. This democratization of the search configuration process was particularly valuable for fine-tuning Bildhistoria’s search experience.

**Fast synchronization:** Bildhistoria required efficient update mechanisms to keep Meilisearch indexes in sync with their database changes. Meilisearch's ability to handle rapid updates and re-indexing was crucial for maintaining an up-to-date and responsive search experience.

**Implementation**
------------------

Bildhistoria's implementation of Meilisearch involved several key steps:

1.  The team initially deployed the open-source version on Azure but faced challenges with the hosting environment.
2.  Bildhistoria team then switched to Meilisearch Cloud, which provided better management and support, allowing them to focus on their core mission rather than infrastructure concerns.
3.  Meilisearch was integrated throughout the platform, powering the main search functionality and an innovative map search feature.
4.  Finally, Bildhistoria invested significant effort in creating an efficient data synchronization pipeline between its database and Meilisearch to handle the constant flow of new information and user contributions.

**Results**
-----------

The implementation of Meilisearch has yielded impressive results for Bildhistoria:

**Multi-index search:** Users can now perform searches that capture results across various types of data—photographs, persons, organizations, and places. This holistic approach has dramatically improved the user experience, allowing for a more comprehensive exploration of historical connections.

**Improved user experience:** The search functionality provides alternative search suggestions for similar queries and provides rich context for each image.

**Interactive map search:** Meilisearch’s [Geosearch](https://meilisearch.com.cn/docs/learn/filtering_and_sorting/geosearch) allows users to explore photographs associated with specific locations in Sweden. As users interact with the map, new searches are performed in real-time, updating the displayed photographs.

![](https://unable-actionable-car.media.strapiapp.com/Bildhistoria_Mapseaechdemogif_ezgif_com_video_to_gif_converter_a70a28b422.gif)

Overall, the team noticed a significant improvement in engagement and user experience since the implementation of Meilisearch.

**Conclusion: Meilisearch is a key part of Bildhistoria's vision for the future**
---------------------------------------------------------------------------------

Looking to the future, Bildhistoria has ambitious plans to further leverage Meilisearch to expand its platform's capabilities to become Sweden's leading platform for historical photographs. While initially focused on the Swedish market, Bildhistoria is open to European collaborations to scale the concept internationally. The next steps for improving the user experience on the platform would include:

*   Integrate AI-driven facial recognition across photographs to link images of the same individuals across different time periods and locations.
*   Add a _houses and buildings_ index, enabling users to trace the lifecycle of specific structures through photographs.
*   Implement features encouraging community collaboration, such as allowing premium users to follow and receive notifications about new edits and contributions.
*   Scale operations to manage a vast archive of 1-3 million photographs by collaborating with more museums and institutions.

As the project progresses and reaches full production mode, Bildhistoria continues to assemble the right technical team to establish a sustainable and long-term setup for the community, laying the groundwork for its future growth and impact.

Meilisearch powers the discovery of history for Bildhistoria

📅 Save the date: October 9th
-----------------------------

[![What's new in Meilisearch: roadmap, v1.10, federated search, and Cloud news!](https://unable-actionable-car.media.strapiapp.com/October_9th_2_f41b190f00.png)](https://us06web.zoom.us/meeting/register/tZUpcu2gqT4pHtcB0Wh8P-aIUd3e4LF-T-UX#/registration)

Join us on October 9th for a **Meilisearch webinar** featuring:

*   A **live demo** of Meilisearch Cloud's latest capabilities
*   A sneak peek at our **product roadmap** and vision
*   Insights on **federated search**
*   **Q&A session** with our experts

We look forward to seeing you there!

<Callout>
  Wednesday, October 9th  
4PM CEST | 3PM BST | 10AM EDT
</Callout>

[Register now](https://us06web.zoom.us/meeting/register/tZUpcu2gqT4pHtcB0Wh8P-aIUd3e4LF-T-UX#/registration)

👀 Meilisearch 1.11 pre-release
-------------------------------

[![Release candidate - Meilisearch 1.11](https://unable-actionable-car.media.strapiapp.com/146_1_12d5d22c9f.jpg)](https://github.com/meilisearch/meilisearch/releases/tag/v1.11.0-rc.0)

Meilisearch 1.11 RC: Major improvements to **hybrid search performance** and usability, plus enhanced **federated search**

❗️Not production-ready. Your feedback is crucial!

[Preview updates](https://github.com/meilisearch/meilisearch/releases/tag/v1.11.0-rc.0)

🤔 Choosing the best model for semantic search
----------------------------------------------

[![](https://unable-actionable-car.media.strapiapp.com/best_sematic_model_8490bd88d2.png)](/blog/choosing-the-best-model-for-semantic-search/#newsletter&utm_source=blog&utm_medium=september-24-updates)

A **comparison** of model performance, cost, and relevancy.

[Read guide](/blog/choosing-the-best-model-for-semantic-search/#newsletter&utm_source=blog&utm_medium=september-24-updates)

🎉 Join Meilisearch's referral program
--------------------------------------

[![New referral program](https://unable-actionable-car.media.strapiapp.com/Referral_program_804eb40f12.png)](https://help.meilisearch.com/en/article/join-meilisearch-referral-program-and-earn-rewards-t3m5rl/#newsletter&utm_source=blog&utm_medium=september-24-updates)

Earn **rewards** while helping grow the community! 🤝 It's a **limited-time** program—don't miss it!

[Learn more](https://help.meilisearch.com/en/article/join-meilisearch-referral-program-and-earn-rewards-t3m5rl/#newsletter&utm_source=blog&utm_medium=september-24-updates)

📰 In the news
--------------

💡 [What is federated search?](/blog/what-is-federated-search/#newsletter&utm_source=blog&utm_medium=september-24-updates)

Learn how it surpasses multi-index search, and elevates both user experience and developer efficiency.

🎮 [Meilisearch playground](https://playground.meilisearch.com/#newsletter&utm_source=blog&utm_medium=september-24-updates)

Explore and compare Meilisearch with other search engines. Test multiple search modes and AI models and evaluate their performance!

📣 [RFC: Similar Documents API](https://github.com/meilisearch/meilisearch/issues/4736)

Join the discussion to help us shape the future of the similar  documents API in Meilisearch.

📣 [RFC: Improving the search preview](https://github.com/orgs/meilisearch/discussions/766)

We want your feedback: how can we make it easier to test your search settings?

✨ [Semantic search with Mistral embeddings](https://meilisearch.com.cn/docs/guides/ai/mistral#newsletter&utm_source=blog&utm_medium=september-24-updates)

Learn how to set up Meilisearch with Mistral embeddings to enhance your search experience and retrieve more relevant results.

🤗 Made with Meilisearch
------------------------

Discover how [Bookshop.org](https://bookshop.org/) increased search-bases purchases by 43% after switching to Meilisearch. 

[![Bookshop.org success story, +43% search-based purchases with Meilisearch](https://unable-actionable-car.media.strapiapp.com/59_fc6540f28f.png)](/blog/bookshop-increases-search-based-purchases/#newsletter&utm_source=blog&utm_medium=september-24-updates)

[Read case study](/blog/bookshop-increases-search-based-purchases/#newsletter&utm_source=blog&utm_medium=september-24-updates)

_We’re always looking to promote new projects from the community. Post yours in the_ [_#built-with-meilisearch_](https://discord.gg/uB94t8PGV4) _channel!_

🦸 We're hiring!
----------------

See our open positions:

*   [Senior Full-Stack Software Engineer](https://meilisearch.com.cn/careers#newsletter&utm_source=blog&utm_medium=september-24-updates)
*   [Founding Sales & Business Developer](https://meilisearch.com.cn/careers#newsletter&utm_source=blog&utm_medium=september-24-updates)  
    

* * *

Stay in the loop of everything Meilisearch by subscribing to our [monthly newsletter](https://dub.sh/meili-newsletter#newsletter&utm_source=blog&utm_medium=september-24-updates). To learn more about Meilisearch's future and help shape it, take a look at our [roadmap](https://roadmap.meilisearch.com/#newsletter&utm_source=blog&utm_medium=september-24-updates) and participate in our [Product Discussions](https://github.com/orgs/meilisearch/discussions#blog.meilisearch.com).

For anything else, join our developers community on [Discord](https://discord.meilisearch.com/#newsletter&utm_source=blog&utm_medium=september-24-updates).

Meilisearch September updates

In today's interconnected world, many applications need to handle content in multiple languages. Meilisearch offers powerful capabilities for multilingual search, combining intelligent language handling with specialized models for optimal relevance. This guide will walk you through setting up multilingual search with Meilisearch, covering both full-text and semantic search approaches.

Why multilingual search matters
===============================

Before we dive into the technical details, let's consider why multilingual search is crucial:

1.  Global reach: Expand your audience by catering to users in their native languages.
2.  Improved user experience: Provide relevant results regardless of the language used in queries or content.
3.  Increased engagement: Users are more likely to interact with content they can easily find and understand.

Now, let's explore how Meilisearch makes multilingual search a breeze.

Full-text search: intelligent language handling
-----------------------------------------------

At its core, Meilisearch excels at handling diverse languages out of the box. Our [advanced tokenizer](https://meilisearch.com.cn/docs/learn/indexing/tokenization#vector-search&utm_source=blog&utm_medium=multilingual_search) is designed to work with:

*   Languages that use whitespace to separate words (like English, Spanish, and French)
*   Languages with different writing systems (such as Chinese, Japanese, and Thai)

### Automatic language detection

When you add documents to your Meilisearch index, the engine automatically detects the language of each field. This means you don't need to specify languages manually in most cases - Meilisearch will handle it for you.

For example, if your documents contain a mix of English and Spanish content:

```json
[
  {
    "id": 1,
    "title": "The quick brown fox",
    "description": "El zorro marrón rápido salta sobre el perro perezoso"
  },
  {
    "id": 2,
    "title": "La luna y las estrellas",
    "description": "A beautiful night sky filled with twinkling stars"
  }
]
```

Meilisearch will automatically detect and properly handle the different languages in each field, ensuring accurate search results across your multilingual content.

### Specifying languages for specific fields

While automatic detection works well in most cases, you might want more control over language handling for certain fields. Meilisearch allows you to [specify languages](https://meilisearch.com.cn/docs/reference/api/settings#localized-attributes#vector-search&utm_source=blog&utm_medium=multilingual_search) for particular attributes using the `localizedAttributes` setting.

Here's how you can set it up:

```json
{
  "localizedAttributes": [
    {
      "locales": ["eng"],
      "attributePatterns": ["title_en", "*_english"]
    },
    {
      "locales": ["spa"],
      "attributePatterns": ["description_es", "*_spanish"]
    }
  ]
}
```

This configuration tells Meilisearch to treat fields ending with `_en` or `_english` as English, and fields ending with `_es` or `_spanish` as Spanish. This approach is particularly useful when you have a consistent naming convention for your multilingual fields or when you want to ensure specific treatment for certain attributes.

Semantic search: Enhancing multilingual relevance
=================================================

While full-text search works well for many multilingual scenarios, semantic search can provide even better results, especially for complex queries or when dealing with nuanced meanings across languages.

Meilisearch supports various embedding models for semantic search, offering powerful auto-embedding capabilities. This means you can easily set up semantic search without having to manage the embedding process yourself. Meilisearch handles the generation and storage of embeddings automatically, seamlessly integrating them into your search experience.

For multilingual applications, we recommend using specialized multilingual models rather than relying on general-purpose models. These models are designed to understand and represent meaning across multiple languages, leading to more relevant search results in multilingual contexts.

One excellent option for multilingual semantic search is the [Cohere Multilingual embedding model](https://docs.cohere.com/docs/multilingual-language-models). It offers a great balance of speed and relevance across a wide range of languages. Cohere's model is particularly adept at understanding context and nuance in different languages, making it an ideal choice for diverse, multilingual content. Follow [the guide](https://meilisearch.com.cn/docs/guides/ai/cohere#vector-search&utm_source=blog&utm_medium=multilingual_search) if you want to do it now.

To explore different models and their performance, you can use the [Meilisearch playground](https://playground.meilisearch.com). This interactive tool allows you to test various embedding models and see how they perform with your specific multilingual queries and content.

Conclusion
----------

We encourage you to experiment with different settings and models to find the optimal configuration for your specific use case. And remember, the Meilisearch community is always here to help - join us on [Discord](https://discord.meilisearch.com/) or check out our documentation for more detailed information and support.

Ready to get started with multilingual search? Try Meilisearch today [with a 14 day trial](https://cloud.meilisearch.com/#vector-search&utm_source=blog&utm_medium=multilingual_search)!

Meilisearch 最新消息和公司动态

Meilisearch 使用二值量化将嵌入索引速度提高7倍

Meilisearch 2024 年第四季度发布总结

Meilisearch 十月更新

TutKit 与 Meilisearch 的旅程：大规模支持多语言学习。

Meilisearch 1.11

Meilisearch 助力 Bildhistoria 探索历史

Meilisearch 9月更新

跨语言搜索

如何向 React 应用添加 AI 驱动的搜索